NWaves Provides methods for conversion between PCM bytes and float[] data. Converts Pcm_8bit to arrays of . Array of bytes Arrays of floats Normalize samples Converts arrays of to Pcm_8bit . Array of bytes Arrays of floats True if samples are normalized Converts Pcm_16bit to arrays of (little-endian or big-endian). Array of bytes Arrays of floats Normalize samples True if Big-endian Converts arrays of to Pcm_16bit (little-endian or big-endian). Array of bytes Arrays of floats True if samples are normalized True if Big-endian Defines channel indexing schemes (used for addressing signals inside audio containers). Left channel ( = 0). Right channel ( = 1). Mono as sum of all channels. Mono as average from all channels. Interleaved channels. Interface for audio containers. Gets the list of discrete signals in container. Gets the signal from container using indexing scheme based on channel type. Channel (left, right, interleave, sum, average, or ordinary index) Interface for audio players. Gets or sets sound volume (usually in range [0..1]). Plays samples contained in asynchronously. Signal to play Index of the first sample to play Index of the last sample to play Number of bits per one sample Plays samples contained in WAV file (or some other source) asynchronously. Path to WAV file (or other source) to play Index of the first sample to play Index of the last sample to play Pauses playing audio. Resumes playing audio. Stops playing audio. Interface for audio recorders. Starts recording audio with specific settings. Sampling rate Number of channels (1=mono, 2=stereo) Number of bits per sample (8, 16, 24 or 32) Stops recording audio and saves recorded sound to file or any other destination. Path to output file (destination) Provides MCI functions imported from winmm.dll. Audio player based on MCI. MciAudioPlayer works only on Windows, since it uses winmm.dll and MCI commands. MciAudioPlayer lets MCI do all the heavy-lifting with sound playback. It launches MCI command and just awaits for amount of time corresponding to the duration of a given segment. Hidden alias for an MCI waveaudio device. Duration of pause in milliseconds. The exact time when playback was paused. The flag indicating whether audio playback is currently paused. Gets or sets audio volume (measured in percents from the range [0..1]). Plays audio contained in WAV file asynchronously. Path to WAV file to play Index of the first sample to play Index of the last sample to play does not implement this method. Unfortunately, MCI does not provide means for playing audio from buffers in memory. Moreover, since NWaves library is portable, there's even no easy way to write the buffer into temporary file and play it here (it could be a workaround for the problem). Signal to play Index of the first sample to play Index of the last sample to play Number of bits per one sample Pauses playing audio. Resumes playing audio. Stops playing audio and close MCI device. Audio recorder based on MCI. MciAudioRecorder works only with Windows, since it uses winmm.dll and MCI commands. Starts recording audio with specific settings. Sampling rate Number of channels (1=mono, 2=stereo) Number of bits per sample (8, 16, 24 or 32) Stops recording audio and save it to WAV file. Path to output WAV file containing recorded sound Represents PCM WAV container. is essentially a constructor of signals in memory based on data from the WAV stream, and its lifetime is not synchronized with the stream whatsoever. is not intended to be a wrapper around the stream, or to acquire any resource (it doesn't affect the underlying stream). The synonym name of this class could be also WaveContainer. Gets the list of discrete signals in container. Gets WAV header (WAVE format). Supported bit depths. Constructs WAV container by loading signals from . Input stream Normalize samples Constructs WAV container by loading signals from a byte array (i.e. byte content of WAV file). Input array of bytes Normalize samples Constructs WAV container by loading signals from part of a byte array (i.e. byte content of WAV file). Input array of bytes Start position in byte array Normalize samples Reads PCM WAV binary data and fills and structure. Input stream of PCM WAV binary data Normalize samples Constructs WAV container by loading into it collection of with given . Signals to be loaded into container Bit depth Constructs WAV container by loading into it one with given . Signal to be loaded into container Bit depth Returns the contents of PCM WAV container as array of bytes. True if samples are normalized Saves the contents of PCM WAV container to . Output stream True if samples are normalized Gets the signal from container using indexing scheme based on channel type. Examples waveFile[Channels.Left] -> waveFile.Signals[0]
waveFile[Channels.Right] -> waveFile.Signals[1] (if it exists)
waveFile[(Channels)2] -> waveFile.Signals[2] (if it exists)
waveFile[Channels.Average] -> returns channel-averaged (new) signal
waveFile[Channels.Interleave] -> returns interleaved (new) signal Channel (left, right, interleave, sum, average, or ordinary index) PCM WAVE header structure. Audio format (PCM = 1). Number of channels (1 - mono, 2 - stereo). Sampling rate (e.g. 8000 Hz, 11025 Hz, 16000 Hz, 22050 Hz, 44100 Hz). SamplingRate * NumChannels * BitsPerSample / 8. ChannelCount * BitsPerSample / 8. Bit depth (bits per sample) (e.g. 8, 16, 24, 32). Represents AutoWah audio effect (envelope follower + Wah-Wah effect). Gets or sets Q factor (a.k.a. Quality Factor, resonance). Gets or sets minimal LFO frequency (in Hz). Gets or sets maximal LFO frequency (in Hz). Gets or sets attack time (in seconds). Gets or sets release time (in seconds). Sampling rate. Internal envelope follower. Constructs . Sampling rate Minimal LFO frequency (in Hz) Maximal LFO frequency (in Hz) Q factor (a.k.a. Quality Factor, resonance) Attack time (in seconds) Release time (in seconds) Processes one sample. Input sample Resets effect. Abstract class for audio effects. Processes one sample. Input sample Resets effect. Applies effect to entire and returns new processed signal. Signal Filtering method Interface for wet/dry mixers. Gets or sets wet gain. Gets or sets dry gain. Defines wet/dry mixing rules. Base class implementing wet/dry mixing logic. Gets or sets wet gain (by default, 1). Gets or sets dry gain (by default, 0). Sets wet/dry mix (in range [0..1]). Wet/dry mix Mixing rule Sets wet/dry gains in decibels and applies linear mix rule. Wet gain in decibels Dry gain in decibels Represents Bitcrusher (distortion) audio effect. Step is calculated from bit depth. Gets or sets the bit depth (number of bits). Constructs with given . Bit depth (number of bits) Processes one sample. Input sample Resets effect. Represents Chorus audio effect. Gets or sets widths for each voice (max delays in seconds). Gets or sets LFO frequencies for each voice. Chorus voices. Constructs . Sampling rate LFO frequencies for each voice Widths (max delays, in seconds) for each voice Constructs from . Sampling rate LFOs (in the form of signal builders) Widths (max delays, in seconds) for each voice Processes one sample. Input sample Resets effect. Represents Delay audio effect. Internal fractional delay line. Sampling rate. Gets or sets delay (in seconds). Gets or sets feedback parameter. Constructs . Sampling rate Delay (in seconds) Feedback Interpolation mode for fractional delay line Max delay for reserving the size of delay line Processes one sample. Input sample Resets effect. Represents Distortion audio effect. Gets or sets distortion mode (soft/hard clipping, exponential, full/half-wave rectify). Gets or sets input gain (in dB). Gets or sets output gain (in dB). Constructs . Distortion mode Input gain (in dB) Output gain (in dB) Processes one sample. Input sample Resets effect. Represents Echo audio effect. Internal fractional delay line. Sampling rate. Gets or sets delay (in seconds). Gets or sets feedback parameter. Constructs . Sampling rate Delay (in seconds) Feedback Interpolation mode for fractional delay line Max delay for reserving the size of delay line Processes one sample. Input sample Resets effect. Represents Flanger audio effect. Internal fractional delay line. Sampling rate. Gets or sets width (in seconds). Gets or sets LFO frequency (in Hz). Gets or sets LFO signal generator. Gets or sets depth. Gets or sets feedback parameter. Gets or sets Inverted mode flag. Gets or sets interpolation mode. Constructs . Sampling rate LFO frequency (in Hz) Width (in seconds) Depth Feedback Inverted mode Interpolation mode for fractional delay line Max width (in seconds) for reserving the size of delay line Constructs from . Sampling rate LFO signal generator Width (in seconds) Depth Feedback Inverted mode Interpolation mode for fractional delay line Max width (in seconds) for reserving the size of delay line Processes one sample. Input sample Resets effect. Represents effect of morphing (blending) two sound signals. Hop length. Size of FFT for analysis and synthesis. Size of frame overlap. Internal FFT transformer. Window coefficients. Offset in the input delay line. Offset in the output buffer. Constructs . Hop size (hop length, number of samples) FFT size Processes one sample of input signal and one sample of the signal to be mixed. Sample of input signal Sample of the signal to mix with input signal Processes one frame (block). Resets effect. Blends (mixes) entire input with entire signal. Input signal Signal to mix with input signal Processes one sample. This method is not implemented in class. Input sample Represents Phaser audio effect. Gets or sets Q factor (a.k.a. Quality Factor, resonance). Gets or sets LFO frequency (in Hz). Gets or sets minimal LFO frequency (in Hz). Gets or sets maximal LFO frequency (in Hz). Get or sets LFO signal generator. Sampling rate. Notch filter with varying center frequency. Constructs . Sampling rate LFO frequency (in Hz) Minimal LFO frequency (in Hz) Maximal LFO frequency (in Hz) Q factor (a.k.a. Quality Factor, resonance) Constructs from . Sampling rate LFO signal generator Q factor (a.k.a. Quality Factor, resonance) Processes one sample. Input sample Resets effect. Represents offline Pitch Shift audio effect based on one of the available TSM algorithms and linear interpolation. does not implement online processing (method ). Gets or sets pitch shift ratio. Gets or sets time-scale modification algorithm. Gets or sets window size (frame length). Gets or sets hop length. Constructs . Pitch shift ratio Window size (frame length) Hop length Time-scale modification algorithm Applies effect to entire and returns new pitch-shifted signal. Input signal Filtering method Processes one sample. This method is not implemented in class. Input sample Resets effect. Represents pitch shift audio effect based on overlap-add filtering and pitch shifting in frequency domain. Gets or sets pitch shift ratio. Frequency resolution. Array of spectrum magnitudes (at the current step). Array of spectrum phases (at the current step). Array of phases computed at the previous step. Array of new synthesized phases (at the current step). Constructs . Sampling rate Pitch shift ratio FFT size Hop length Processes one spectrum at each Overlap-Add STFT step. Real parts of input spectrum Imaginary parts of input spectrum Real parts of output spectrum Imaginary parts of output spectrum Resets effect. Represents audio effect of speech robotization. Usual settings: fftSize = 512, hopSize = 70 .. 270 fftSize = 1024, hopSize = 140 .. 540 fftSize = 2048, hopSize = 280 .. 1080 etc. Constructs . Hop size (hop length, number of samples) FFT size Processes one spectrum at each Overlap-Add STFT step (simply set phases to 0). Real parts of input spectrum Imaginary parts of input spectrum Real parts of output spectrum Imaginary parts of output spectrum Represents binaural panning audio effect (HRIR/BRIR interpolation + optional crossover filter). Left ear HRIR. Right ear HRIR. HRIRs table (left ear). HRIRs table (right ear). Azimuths (thetas). Elevations (phis). Left ear HRIR convolver. Right ear HRIR convolver. Turn on/off crossover filtering. Crossover filters (low-pass part). Crossover filters (low-pass part). Crossover filters (high-pass part). Crossover filters (high-pass part). Gets or sets azimuth. Gets or sets elevation. Constructs . For example (CIPIC): 25 azimuths (theta):
[-80 -65 -55 -45 -40 -35 -30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 55 65 80]

50 elevations (phi):
[-45 -39 -34 -28 -23 -17 -11 -6 0 6 11 17 23 28 34 39 45 51 56 62 68 73 79 84 90 96
101 107 113 118 124 129 135 141 146 152 158 163 169 174 180 186 191 197 203 208 214 219 225 231]
Azimuths (thetas) - must be sorted in ascending order Elevations (phis) - must be sorted in ascending order HRIR collection (left ear) HRIR collection (right ear) Turns on/off crossover filtering. Updates frequency of the crossover filter (works only for BiQuadFilters). Filters of other types / parameters can be passed to constructor. Frequency Sampling rate Sets custom crossover filters. Crossover filter (low-pass part) for left channel Crossover filter (high-pass part) for left channel Crossover filter (low-pass part) for right channel Crossover filter (high-pass part) for right channel Updates HRIR (interpolates it using HRIR tables). Azimuth (theta) Elevation (phi) Processes one sample in each of two channels : [ input left , input right ] -> [ output left , output right ]. Input sample in left channel Input sample in right channel Resets binaural pan effect. Represents stereo pan audio effect based on ITD-ILD (Interaural Time Difference - Interaural Level Difference). Speed of sound. Constant pi/2. Gets head radius. Sampling rate. Head factor. ITD delay lines. ITD delay lines. ILD filters. ILD filters. Time delays (in seconds). Time delays (in seconds). Gets or sets pan. Interaural Time Difference. Angle Constructs . Sampling rate Pan Interpolation mode for fractional delay line Max delay (in seconds) for reserving the size of delay line Head radius Processes one sample in each of two channels : [ input left , input right ] -> [ output left , output right ]. Input sample in left channel Input sample in right channel Resets effect. Represents stereo panning audio effect. Pan value (must be in range [-1..1]). Pan value for calculations (in range [0..1]). Pan value for calculations in constant-power mode (in range [0..1]). Gets or sets pan (in range [-1..1]). Gets or sets pan rule (pan law). Constructs . Pan Pan rule (pan law) Processes one sample in each of two channels : [ input left , input right ] -> [ output left , output right ]. Input sample in left channel Input sample in right channel Resets effect. Defines pan rules (pan laws). Represents stereo ping-pong delay audio effect. Left channel delay line. Righ channel delay line. Sampling rate. Gets or sets pan. Gets or sets delay (in seconds). Gets or sets feedback coefficient. Constructs . Sampling rate Pan Delay (in seconds) Feedback Interpolation mode for fractional delay line Max delay for reserving the size of delay line Processes one sample in each of two channels : [ input left , input right ] -> [ output left , output right ]. Input sample in left channel Input sample in right channel Resets effect. Represents stereo delay audio effect. Left channel delay effect. Right channel delay effect. Gets or sets left channel delay (in seconds). Gets or sets right channel delay (in seconds). Gets or sets left channel feedback. Gets or sets right channel feedback. Gets or sets pan. Constructs . Sampling rate Pan Left channel delay (in seconds) Left channel feedback Right channel delay (in seconds) Right channel feedback Interpolation mode for fractional delay line Max delay for reserving the size of delay line Processes one sample in each of two channels : [ input left , input right ] -> [ output left , output right ]. Input sample in left channel Input sample in right channel Resets effect. Abstract class for stereo audio effects. Processes one sample in each of two channels : [ input left , input right ] -> [ output left , output right ]. Input sample in left channel Input sample in right channel Processes one sample in mono channel : input sample -> [ output left , output right ]. Input sample in mono channel Output sample for left channel Output sample for right channel Processes blocks of samples in each of two channels : [ input left , input right ] -> [ output left , output right ]. Input block of samples (left channel) Input block of samples (right channel) Output block of samples (left channel) Output block of samples (right channel) Number of samples to process Input starting index Output starting index Processes block of samples in mono channel : [ input ] -> [ output left , output right ]. Input block of samples (mono channel) Output block of samples (left channel) Output block of samples (right channel) Number of samples to process Input starting index Output starting index Applies effect to entire and returns tuple of output signals [left signal, right signal]. Input signal Applies effect to entire signals (in left and right channels) and returns tuple of output signals [left signal, right signal]. Input signal (left channel) Input signal (right channel) Resets effect. Represents Tremolo audio effect. Gets or sets depth. Gets or sets tremolo frequency (modulation frequency) (in Hz). Gets or sets tremolo index (modulation index). Gets or sets LFO signal generator. Constructs . Sampling rate Depth Tremolo frequency (modulation frequency) (in Hz) Tremolo index (modulation index) Constructs from . LFO signal generator Depth Processes one sample. Input sample Resets effect. Represents Tube Distortion audio effect. Gets or sets input gain (in dB). Gets or sets output gain (in dB). Gets or sets Q factor (Work point). Controls the linearity of the transfer function for low input levels. More negative - more linear. Gets or sets distortion's character. Higher number - harder distortion. Gets filter coefficient (close to 1.0) defining placement of poles in the HP filter that removes DC component. Gets filter coefficient (in range [0, 1]) defining placement of pole in the LP filter used to simulate capacitances in tube amplifier. Internal filter for output signal that combines HP and LP filters mentioned above. Constructs . Input gain (in dB) Output gain (in dB) Q factor (controls the linearity of the transfer function for low input levels. More negative means more linear) Distortion's character (higher number means harder distortion) Filter coefficient (close to 1.0) defining placement of poles in the HP filter that removes DC component Filter coefficient (in range [0, 1]) defining placement of pole in the LP filter used to simulate capacitances in tube amplifier Processes one sample. Input sample Resets effect. Represents Vibrato audio effect. Internal fractional delay line. Sampling rate. Gets or sets width (in seconds). Gets or sets LFO frequency (in Hz). Gets or sets LFO signal generator. Gets or sets interpolation mode. Constructs . Sampling rate LFO frequency (in Hz) Width (in seconds) Interpolation mode for fractional delay line Max width (in seconds) for reserving the size of delay line Constructs from . Sampling rate LFO signal generator Width (in seconds) Interpolation mode for fractional delay line Max width (in seconds) for reserving the size of delay line Processes one sample. Input sample Resets effect. Represents Wah-Wah audio effect. Gets or sets LFO frequency (in Hz). Gets or sets minimal LFO frequency (in Hz). Gets or sets maximal LFO frequency (in Hz). Gets or sets Q factor (a.k.a. Quality Factor, resonance). Gets or sets LFO signal generator. Sampling rate. Constructs . Sampling rate LFO frequency (in Hz) Minimal LFO frequency (in Hz) Maximal LFO frequency (in Hz) Q factor (a.k.a. Quality Factor, resonance) Constructs from . Sampling rate LFO signal generator Q factor (a.k.a. Quality Factor, resonance) Processes one sample. Input sample Resets effect. Represents audio effect of speech whisperization. Hint. Choose relatively small FFT and hop sizes (e.g., 256 and 40). Phase randomizer. Constructs . Hop size (hop length, number of samples) FFT size Processes one spectrum at each Overlap-Add STFT step. Real parts of input spectrum Imaginary parts of input spectrum Real parts of output spectrum Imaginary parts of output spectrum Represents amplitude modulation spectra (AMS) extractor. Gets feature annotations. They are initialized in constructor in the following manner (example): band_1_mf_0.5_Hz band_1_mf_1.0_Hz ... band_1_mf_8.0_Hz
band_2_mf_0.5_Hz band_2_mf_1.0_Hz ... band_2_mf_8.0_Hz
...
band_32_mf_0.5_Hz band_32_mf_1.0_Hz ... band_32_mf_8.0_Hz
The "featuregram": the sequence of feature vectors. If this sequence is given, then computes modulation spectral coefficients from sequences in each 'feature channel'. Filterbank matrix of dimension [filterCount * (fftSize/2 + 1)]. Gets filterbank matrix of dimension [filterCount * (fftSize/2 + 1)]. Signal envelopes in different frequency bands. Gets signal envelopes in different frequency bands. Size of FFT. FFT transformer. FFT transformer for modulation spectrum. Size of FFT applied to signal envelopes. Hop size for analysis of signal envelopes. Internal buffer for a signal block at each step. Internal buffer for a signal spectrum at each step. Internal buffer for filtered spectrum. Internal buffer for modulation spectrum analysis. Modulation spectrum (in one band). Constructs extractor from configuration . Computes modulation spectra. Each vector representing one modulation spectrum is a flattened version of 2D spectrum. Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Creates 2D modulation spectrum from its flattened version. Axes are: [short-time-frequency] x [modulation-frequency]. AMS feature vector Gets sequence of short-time spectra corresponding to particular modulation frequency (by default, the most perceptually important modulation frequency of 4 Hz). Sequence of AMS feature vectors Modulation frequency does not provide this function. Call method instead. Block of data Features (one feature vector) computed in the block does not provide this function. Call overloaded method . Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Pre-allocated sequence for storing the resulting feature vectors Abstract class for all NWaves-style feature extractors. Gets number of features to extract (feature vector size). Gets string annotations (or simply names) of features. Gets string annotations (or simply names) of delta features (1st order derivatives). Gets string annotations (or simply names) of delta-delta features (2nd order derivatives). Gets length of analysis frame (duration in seconds). Gets hop length (duration in seconds). Gets size of analysis frame (number of samples). Gets hop size (number of samples). Gets expected sampling rate of signals for analysis. Size of the block for processing at each step. This field is usually set in subclass methods. Pre-emphasis coefficient. Type of the window function. Window samples. Constructs extractor from configuration . Computes feature vectors from and stores them in . Returns the number of actually computed feature vectors Array of samples Index of the first sample in array for processing Index of the last sample (exclusive) in array for processing Pre-allocated sequence for storing the resulting feature vectors Computes feature vectors from and stores them in . Returns the number of actually computed feature vectors Array of samples Pre-allocated sequence for storing the resulting feature vectors Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Index of the first sample in array for processing Index of the last sample (exclusive) in array for processing Returns time markers (in seconds). Number of feature vectors Starting time position (in seconds) Processes one frame in block of data at each step (in general block can be longer than frame, e.g. zero-padded block for FFT). Block of data Features (one feature vector) computed in the block Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the signal length is less than the size of analysis frame. Discrete signal Index of the first sample in signal for processing Index of the last sample (exclusive) in signal for processing Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the signal length is less than the size of analysis frame. Discrete signal Resets feature extractor. Returns true if the extractor supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Returns null if the extractor does not support parallelization. Computes parallelly feature vectors from . Returns chunks of fecture vector lists computed in each separate thread. Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Number of threads (all available processors, by default) Computes parallelly feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Number of threads (all available processors, by default) Computes parallelly feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Number of threads (all available processors, by default) Computes parallelly feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Discrete signal Index of the first sample in signal for processing Index of the last sample in signal for processing Number of threads (all available processors, by default) Computes parallelly feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Discrete signal Number of threads (all available processors, by default) Provides methods for post-processing of feature vector sequences. Does mean subtraction (in particular, CMN). Sequence of feature vectors Does variance normalization (division by estimate of std.deviation (biased or not biased)). Sequence of feature vectors Bias in estimate of variance (1 = not biased, 0 = biased) Extends feature vectors with delta-features (1st and optionally 2nd order derivatives). According to formula: d_t = \frac{\sum_{n=1}^N n (c_{t+n} - c_{t-n})}{2 \sum_{n=1}^N n^2} vectors should be introduced before and after input vectors for computations.
By default, these vectors will be created automatically and copied from the first and the last input vector.
They can also be specified explicitly in and parameters. Sequence of feature vectors that will be extended Sequence of feature vectors that will be prepended to for computations Sequence of feature vectors that will be appended to for computations Should delta-delta features be computed Number of feature vectors before and after input for computations Joins (merges) feature vectors from different collections into one combined feature vector. For example, it can join 12 MFCC and 10 PLP coeffs into one 22-dimensional vector. Collections of feature vectors must have the same size and contain at least one vector. Sequences of feature vectors Provides extension methods for feature vectors. Creates dictionary with feature vector statistics (keys are: min, max, mean, var). Feature vector Interface for feature extractors. Gets number of features to extract (feature vector size). Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Computes feature vectors from and stores them in . Returns the number of actually computed feature vectors Array of samples Pre-allocated sequence for storing the resulting feature vectors Computes feature vectors from and stores them in . Returns the number of actually computed feature vectors Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Pre-allocated sequence for storing the resulting feature vectors Resets feature extractor. Interface for feature extractors that support parallelized computations. Computes parallelly the feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Number of threads Computes parallelly the feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Number of threads adapter for online feature extraction. Gets or sets underlying feature extractor. Gets number of features to extract (feature vector size). Should the last non-processed samples in the current block be ignored in the next block. The number of last non-processed samples in the current block that will be processed in the next block. Internal buffer for accumulated samples. Constructs as a wrapper around . Underlying feature extractor Should the last non-processed samples in the current block be ignored in the next block Reserved max size of the internal buffer for accumulated samples Returns maximally possible number of output feature vectors (based on maximally possible online data portion size). This number is intended to be used for pre-allocation of feature vector lists. Maximally possible online data portion size (number of samples) Returns maximally possible number of output feature vectors (based on maximally possible duration of online data portion). This number is intended to be used for pre-allocation of feature vector lists. Maximally possible duration of online data portion (in seconds) Ensures the size of internal buffer for accumulated samples. If the new size exceeds the buffer size, it will be auto-resized. Required size (measured in sample count) Ensures the size of internal buffer for accumulated samples based on required duration. If the new size (computed from duration) exceeds the buffer size, it will be auto-resized. Required duration in seconds Computes feature vectors from and stores them in . Returns the number of actually computed feature vectors. Block of data Pre-allocated sequence for storing the resulting feature vectors Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Block of data Computes feature vectors from and stores them in . Returns the number of actually computed feature vectors. Block of data Index of the first sample in array for processing Index of the last sample in array for processing Pre-allocated sequence for storing the resulting feature vectors Computes feature vectors from . Returns the list of computed feature vectors or empty list, if the number of samples is less than the size of analysis frame. Block of data Index of the first sample in array for processing Index of the last sample in array for processing Resets online feature extractor. Represents chroma features extractor. Gets feature names: "C", "C#", "D", "D#", etc. if chroma count == 12 and baseC == true; "A", "A#", "B", "C", etc. if chroma count == 12 and baseC == false; "chroma1", "chroma2", etc. otherwise. Filterbank matrix of dimension [ChromaCount * (blockSize/2 + 1)]. Gets filterbank matrix of dimension [ChromaCount * (blockSize/2 + 1)]. FFT transformer. Internal buffer for a signal spectrum at each step. Chroma extractor options. Constructs extractor from configuration . Computes chroma feature vector in one frame. Block of data Features (one chroma feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. computes in each frame spectral energies in frequency bands defined by a given filterbank (channel outputs). So it's like MFCC but without DCT-compressing of the filterbank-mapped spectrum. Gets feature names (simply "fb0", "fb1", "fb2", etc.) Gets filterbank matrix of dimension [filterbankSize * (blockSize/2 + 1)]. FFT transformer. Non-linearity type (logE, log10, decibel, cubic root). Spectrum calculation scheme (power/magnitude normalized/not normalized). Floor value for LOG calculations. Delegate for calculating spectrum. Delegate for post-processing spectrum. Internal buffer for a signal spectrum at each step. Internal buffer for a post-processed band spectrum at each step. Constructs extractor from configuration . Computes vector of filter bank channel outputs in one frame. Block of data Features (one feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Represents Linear Prediction Cepstral Coefficients (LPCC) extractor. Gets feature names (simply "lpcc0", "lpcc1", etc.) Order of an LPC-filter. Size of liftering window. Liftering window coefficients. Internal convolver. Internal buffer for cross-correlation signal. Internal buffer for LPC-coefficients. Internal buffer for reversed real parts of the currently processed block. Constructs extractor from configuration . Computes LPCC vector in one frame. Block of data Features (one LPCC feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Represents Linear Predictive Coding (LPC) coefficients extractor. Gets feature names ("error", "lpc1", "lpc2", etc.) Order of an LPC-filter. Internal convolver. Internal buffer for reversed real parts of the currently processed block. Internal buffer for cross-correlation signal. Constructs extractor from configuration . Computes LPC vector in one frame. Note: The first LP coefficient is always equal to 1.0. This method replaces it with the value of prediction error. Block of data Features (one LPC feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Represents Mel Frequency Cepstral Coefficients (MFCC) extractor. Since so many variations of MFCC have been developed since 1980, this class is very general and allows customizing a lot of parameters: filterbank (by default it's MFCC-FB24 HTK/Kaldi-style) non-linearity type (logE, log10, decibel (librosa power_to_db analog), cubic root) spectrum calculation type (power/magnitude normalized/not normalized) DCT type (1,2,3,4 normalized or not): "1", "1N", "2", "2N", etc. floor value for LOG-calculations (usually it's float.Epsilon; HTK default seems to be 1.0 and in librosa 1e-10 is used) Gets feature names (simply "mfcc0", "mfcc1", "mfcc2", etc.) Gets filterbank matrix of dimension [filterbankSize * (fftSize/2 + 1)]. By default it's mel filterbank. Size of liftering window. Liftering window coefficients. FFT transformer. DCT transformer. DCT type ("1", "1N", "2", "2N", "3", "3N", "4", "4N"). Non-linearity type (logE, log10, decibel, cubic root). Spectrum calculation scheme (power/magnitude normalized/not normalized). Floor value for LOG calculations. Should the first MFCC coefficient be replaced with LOG(energy). Floor value for LOG-energy calculation. Delegate for calculating spectrum. Delegate for post-processing spectrum. Delegate for applying DCT. Internal buffer for a signal spectrum at each step. Internal buffer for a post-processed mel-spectrum at each step Constructs extractor from configuration . Computes MFCC feature vector in one frame. General algorithm: Apply window Obtain power spectrum Apply mel filterbank (or any other filterbank) Apply nonlinearity (log, by default) Do DCT [Optional] lifter MFCC-cepstrum Block of data Features (one MFCC feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. follows MPEG-7 recommendations to evaluate the following features: Spectral features (MPEG-7) Harmonic features Perceptual features It's a flexible extractor that allows varying almost everything. The difference between and is that former calculates spectral features from total energy in frequency BANDS while latter analyzes signal energy at particular frequencies (spectral bins). Check FeatureSet and HarmonicSet to see the full list of supported features. Full set of supported spectral features. Full set of supported harmonic features. Gets string annotations (or simply names) of features. Filterbank from frequency bands. Internal buffer for frequency bands. Internal buffer for central frequencies. Internal buffer for harmonic peak frequencies (optional). Internal buffer for spectral positions of harmonic peaks (optional). Extractor functions. Extractor parameters. Harmonic extractor functions (optional). FFT transformer. Internal buffer for magnitude spectrum. Internal buffer for total energies in frequency bands. Pitch estimator function (optional). Array of precomputed pitches (optional). Current position in pitch track. Harmonic peaks detector function (optional). Constructs extractor from configuration . Adds set of harmonic features to extractor's list. is the function that should be used for pitch estimation. By default, method is called. is the function that should be used for peak detection. By default, method is called. String names/annotations of newly added harmonic features Max number of harmonic peaks Function that should be used for pitch estimation Function that should be used for peak detection Lower frequency of expected pitch range Upper frequency of expected pitch range Adds user-defined harmonic feature to extractor's list (and the routine for its calculation). Feature name/annotation Routine for calculation of the feature Sets array of precomputed pitches. Array of pitches computed elsewhere Computes MPEG-7 feature vectors from and stores them in . Returns the number of actually computed feature vectors Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Pre-allocated sequence for storing the resulting feature vectors Computes MPEG-7 features in one frame. Block of data Features (one feature vector) computed in the block Returns false if array of pitches was set manually, since in this case does not support parallelization. Returns true in all other cases. Creates thread-safe copy of the extractor for parallel computations. Returns null if the extractor does not support parallelization. Represents extractor of spectral features. It's a flexible extractor that allows setting frequencies of interest. At least one spectral feature must be specified. Check FeatureSet to see the full list of supported features. Full set of supported spectral features. Gets string annotations (or simply names) of features. Extractor functions. Extractor parameters. FFT transformer. Center frequencies (uniform in Herz scale by default; could be uniform in mel-scale or octave-scale, for example). Internal buffer for magnitude spectrum. Internal buffer for magnitude spectrum taken only at frequencies of interest. Internal buffer for spectral positions of frequencies of interest. Constructs extractor from configuration . Adds user-defined spectral feature to extractor's list (and the routine for its calculation). Feature name/annotation Routine for calculation of the feature Computes spectral features in one frame. Block of data Features (one feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Represents extractor of time-domain features (energy, rms, ZCR, entropy). Full set of features. Gets string annotations (or simply names) of features. Extractor functions. Parameters. Constructs extractor from configuration . Adds user-defined feature to extractor's list (and the routine for its calculation). Feature name/annotation Routine for calculation of the feature Computes feature vectors from and stores them in . Returns the number of actually computed feature vectors. Array of samples Index of the first sample in array for processing Index of the last sample in array for processing Pre-allocated sequence for storing the resulting feature vectors Processes one frame in block of data at each step. does not provide this function. Call method instead. Block of data Features (one feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Rectangular ModulationFftSize = 64 ModulationHopSize = 4 Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: Number of chroma coefficients must be positive Norm must be positive Octave width must be positive Default values: FrameDuration = 0.025 HopDuration = 0.01 FeatureCount = 12 Window = WindowType.Hann OctaveWidth = 2 CenterOctave = 5.0 Norm = 2 BaseC = true Defines basic properties for configuring feature extractors. General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Rectangular Gets or sets number of features to extract (feature vector size). Gets or sets expected sampling rate of signals for analysis. Gets or sets length of analysis frame (duration in seconds). By default, 0.025 (25ms). Gets or sets hop length (duration in seconds). By default, 0.01 (10ms). Gets or sets size of analysis frame (number of samples). Has priority over . If it is not specified explicitly, then it is auto-computed from . Gets or sets hop size (number of samples). Has priority over . If it is not specified explicitly, then it is auto-computed from . Gets or sets pre-emphasis filter coefficient. Gets or sets window function (by default, rectangular window, i.e. no windowing). In and its subclasses (MFCC, PLP, etc.) default window is Hamming. Returns the list of error messages describing particular configuration validation problem. Provides methods for serializing and casting feature extractor configuration options. Serializes feature extractor configuration options to JSON. Output stream for JSON data Feature extractor configuration options Deserializes feature extractor configuration options from JSON. Options type Input stream containing JSON data Casts feature extractor configuration options of one type to options of another type. Original options type Required options type Feature extractor configuration options Defines properties for configuring filterbank-based extractors (including MFCC, PLP, etc.). General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: Filter bank size must be positive number or the entire filter bank must be specified (not null) Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Hamming FilterbankSize = 12 (acts as FeatureCount) LowFrequency = 0 HighFrequency = 0 (i.e. it will be auto-computed as SamplingRate/2) NonLinearity = NonLineriatyType.None SpectrumType = SpectrumType.Power LogFloor = 1.4e-45 (float epsilon) Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: FeatureCount must be positive mumber LPC order must be positive mumber (usually, FeatureCount-1) Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Rectangular LifterSize = 22 Gets or sets number of liftered coefficients (by default, 22). Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: LPC order must be positive mumber Default values: FrameDuration = 0.025 HopDuration = 0.01 FeatureCount = LPC order + 1 (auto-computed) Window = WindowType.Rectangular Gets or sets order of LPC. This property is required. It has priority over FeatureCount. FeatureCount will be autocomputed by as LpcOrder+1. Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: FeatureCount must be positive mumber DCT type must have one of these values: "1", "2", "3", "4", "1N", "2N", "3N", "4N" Filter bank size must be positive number or the entire filter bank must be specified (not null) FeatureCount must not exceed filterbank size Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Hamming FilterbankSize = 24 LowFrequency = 0 HighFrequency = 0 (i.e. it will be auto-computed as SamplingRate/2) DctType = "2N" NonLinearity = NonLineriatyType.Log10 SpectrumType = SpectrumType.Power LifterSize = 0 (no liftering) LogFloor = 1.4e-45 (float epsilon) LogEnergyFloor = 1.4e-45 (float epsilon) Gets or sets number of liftered coefficients (0, by default, i.e. there will be no liftering). Gets or sets DCT type (by default, it's normalized DCT-II, or "2N"). Defines properties for configuring multi-feature extractors. General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Rectangular FeatureList = "all" Defines non-linear post-processing schemes. Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: HighFrequency must be greater than LowFrequency Default values: LowFrequency = 80 (Hz) HighFrequency = 400 (Hz) Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: FeatureCount must be positive mumber Filter bank size must be positive number or the entire filter bank must be specified (not null) Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Hamming FilterbankSize = 24 LowFrequency = 0 HighFrequency = 0 (i.e. it will be auto-computed as SamplingRate/2) LpcOrder = 0 (i.e. it will be autocomputed as FeatureCount-1) NonLinearity = NonLineriatyType.Log10 SpectrumType = SpectrumType.Power LifterSize = 0 (no liftering) Rasta = 0 (no RASTA-filtering) LogFloor = 1.4e-45 (float epsilon) LogEnergyFloor = 1.4e-45 (float epsilon) Gets or sets order of LPC (0, by default, i.e. it will be autocomputed as FeatureCount-1). Gets or sets coefficient of RASTA-filter (0, by default, i.e. there will be no RASTA-filtering). Gets or sets number of liftered coefficients (0, by default, i.e. there will be no liftering). Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Specific contracts are: FeatureCount must be positive mumber Filter bank size must be positive number or the entire filter bank must be specified (not null) Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Hamming FilterbankSize = 40 LowFrequency = 100 HighFrequency = 6800 Power = 15 SpectrumType = SpectrumType.Power LifterSize = 0 (no liftering) LogFloor = 1.4e-45 (float epsilon) LogEnergyFloor = 1.4e-45 (float epsilon) Defines spectrum calculation modes. Sqrt(re*re + im*im). re*re + im*im. Sqrt(re*re + im*im) / fftSize. (re*re + im*im) / fftSize. Defines properties for configuring . General contracts are: Sampling rate must be positive number Frame duration must be positive number Hop duration must be positive number Default values: FrameDuration = 0.025 HopDuration = 0.01 Window = WindowType.Rectangular WaveletName = "haar" Represents pitch extractor / tracker. Pitch extractor calls autocorrelation method since it's best in terms of universality and quality. The feature vector contains 1 component : pitch. If there's a need to create pitch extractor based on other time-domain method (YIN or ZcrSchmitt), then can be used. If there's a need to create pitch extractor based on a certain spectral method (HSS or HPS), then can be used. Example: var extractor = new TimeDomainFeaturesExtractor(sr, "en", 0.0256, 0.010);
extractor.AddFeature("yin", (s, start, end) => { return Pitch.FromYin(s, start, end); });
var pitches = extractor.ComputeFrom(signal); Gets names of pitch estimation algorithms. Lower frequency of expected pitch range. Upper frequency of expected pitch range. Internal convolver. Internal buffer for reversed real parts of the currently processed block. Internal buffer for cross-correlation signal. Constructs extractor from configuration . Computes pitch in one frame. Block of data Pitch (feature vector containing only pitch) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Represents Perceptual Linear Predictive Coefficients extractor (PLP-RASTA). Gets feature names (simply "plp0", "plp1", "plp2", etc.) Gets filterbank matrix of dimension [filterbankSize * (fftSize/2 + 1)]. By default it's bark filterbank like in original H.Hermansky's work (although many specialists prefer mel bands). Filterbank center frequencies. RASTA coefficient (if zero, then no RASTA filtering). RASTA filters for each critical band. Size of liftering window. Liftering window coefficients. Should the first PLP coefficient be replaced with LOG(energy). Floor value for LOG-energy calculation. FFT transformer. Internal buffer for a signal spectrum at each step. Internal buffer for a signal spectrum grouped to frequency bands. Equal loudness weighting coefficients. LPC order. Internal buffer for LPC-coefficients. Precomputed IDFT table. Autocorrelation samples (computed as IDFT of power spectrum). Constructs extractor from configuration . Computes PLP-RASTA feature vector in one frame. General algorithm: Apply window Obtain power spectrum Apply filterbank of bark bands (or mel bands) [Optional] filter each component of the processed spectrum with a RASTA filter Apply equal loudness curve Apply nonlinearity (take cubic root) Do LPC Convert LPC to cepstrum [Optional] lifter cepstrum Block of data Features (one PLP feature vector) computed in the block Resets extractor. Returns false in RASTA-filtering mode (i.e. if RASTA-coefficient is not 0), since in this case does not support parallelization. Returns true in all other cases. Creates thread-safe copy of the extractor for parallel computations. Returns null if the extractor does not support parallelization. Represents Power-Normalized Cepstral Coefficients (PNCC) extractor. Gets feature names (simply "pncc0", "pncc1", "pncc2", etc.) Gets or sets window length for median-time power (2 * M + 1). Gets or sets window length for spectral smoothing (2 * N + 1). Gets or sets lambda_a used in asymmetric noise suppression formula (4). Gets or sets lambda_b used in asymmetric noise suppression formula (4). Gets or sets the forgetting factor in temporal masking formula. Gets or sets the forgetting factor in formula (15) in [Kim and Stern, 2016]. Gets or sets threshold for detecting excitation/non-excitation segments. Gets or sets the multiplier in formula (12) in [Kim and Stern, 2016]. Gets filterbank matrix of dimension [filterbankSize * (fftSize/2 + 1)]. By default it's gammatone filterbank. Nonlinearity coefficient (if 0 then Log10 is applied). Should the first PNCC coefficient be replaced with LOG(energy). Floor value for LOG-energy calculation. FFT transformer. DCT-II transformer. Internal buffer for a signal spectrum at each step. Value for mean normalization. Ring buffer for efficient processing of consecutive spectra. Step of PNCC algorithm. Constructs extractor from configuration . Computes PNCC vector in one frame according to [Kim and Stern, 2016]. General algorithm: Apply window Obtain power spectrum Apply gammatone filters (squared) Medium-time processing (asymmetric noise suppression, temporal masking, spectral smoothing) Apply nonlinearity Do DCT-II (normalized) Block of data Features (one PNCC feature vector) computed in the block Resets extractor. Helper Ring Buffer class for efficient processing of consecutive spectra. Constructs with given and . Adds to the ring buffer. Resets ring buffer. Provides functionality of simple CSV serialization of feature vectors. Sequence of feature vectors for serialization. Sequence of time markers for serialization. Sequence of feature names/annotations for serialization. Delimiter symbol. Constructs from the list of feature vectors for serialization. Sequence of feature vectors for serialization Sequence of time markers for serialization Sequence of feature names/annotations for serialization Delimiter char Serializes feature vectors into using for values. Output stream Format/precision of values Format/precision of time markers Represents Simplified Power-Normalized Cepstral Coefficients (SPNCC) extractor. Gets feature names (simply "spncc0", "spncc1", "spncc2", etc.) Gets or sets the forgetting factor in formula (15) in [Kim, Stern, 2016]. Gets filterbank (gammatone by default). Nonlinearity coefficient (if 0 then Log10 is applied). Should the first SPNCC coefficient be replaced with LOG(energy). Floor value for LOG-energy calculation. FFT transformer. DCT-II transformer. Internal buffer for a signal spectrum at each step. Internal buffer for gammatone spectrum. Value for mean normalization. Constructs extractor from configuration . Computes S(implified)PNCC vector in one frame according to [Kim and Stern, 2016]. General algorithm: Apply window Obtain power spectrum Apply gammatone filters (squared) Mean power normalization Apply nonlinearity Do DCT-II (normalized) Block of data Features (one SPNCC feature vector) computed in the block Resets extractor. Represents wavelet extractor. Gets feature names (simply "w0", "w1", etc.) Fast Wavelet Transformer. Wavelet name. FWT level (0 = auto). Internal buffer for FWT coefficients. Constructs extractor from configuration . Computes vector of FWT coefficients in one frame. Block of data Features (one FWT feature vector) computed in the block Returns true, since always supports parallelization. Creates thread-safe copy of the extractor for parallel computations. Provides methods for computing harmonic spectral features. Finds harmonic peaks (peak indices and frequencies) in spectrum. If is not specified explicitly, it will be auto-estimated using method . Spectrum Array for storing computed peak positions Array for storing computed peak frequencies Sampling rate Pitch in Hz, or any negative number if the pitch is unknown Computes harmonic centroid. Spectrum Peak positions (indices in spectrum) Peak frequencies Computes harmonic spread. Spectrum Peak positions (indices in spectrum) Peak frequencies Computes inharmonicity. Spectrum Peak positions (indices in spectrum) Peak frequencies Computes harmonic odd-to-even ratio. Spectrum Peak positions (indices in spectrum) Computes tristimulus (th component). Spectrum Peak positions (indices in spectrum) Tristimulus component index: 1, 2 or 3 Provides methods for computing perceptual audio features. Computes perceptual loudness (the sum of specific loudnesses: N'(z) = E(z)^0.23). Array of energies in given spectral bands Computes perceptual sharpness (essentially, the equivalent of spectral centroid). Array of energies in given spectral bands Provides methods for pitch estimation and tracking. Estimates pitch from using autocorrelation method. Array of samples Sampling rate Index of the first sample in array for processing Index of the last sample in array for processing Lower frequency of expected pitch range Upper frequency of expected pitch range Estimates pitch from using autocorrelation method. Signal Index of the first sample in signal for processing Index of the last sample in signal for processing Lower frequency of expected pitch range Upper frequency of expected pitch range Estimates pitch from based on zero crossing rate and Schmitt trigger. Array of samples Sampling rate Index of the first sample in array for processing Index of the last sample in array for processing Lower threshold in Schmitt trigger Upper threshold in Schmitt trigger Estimates pitch from based on zero crossing rate and Schmitt trigger. Signal Index of the first sample in array for processing Index of the last sample in array for processing Lower threshold in Schmitt trigger Upper threshold in Schmitt trigger Estimates pitch from using YIN algorithm: De Cheveigne, A., Kawahara, H. YIN, a fundamental frequency estimator for speech and music. The Journal of the Acoustical Society of America, 111(4). - 2002. Array of samples Sampling rate Index of the first sample in array for processing Index of the last sample in array for processing Lower frequency of expected pitch range Upper frequency of expected pitch range CMDF threshold Estimates pitch from using YIN algorithm. Signal Index of the first sample in signal for processing Index of the last sample in signal for processing> Lower frequency of expected pitch range Upper frequency of expected pitch range CMDF threshold Estimates pitch from using Harmonic Sum Spectrum (HSS) method. Signal Index of the first sample in signal for processing Index of the last sample in signal for processing Lower frequency of expected pitch range Upper frequency of expected pitch range FFT size Estimates pitch from using Harmonic Sum Spectrum (HSS) method. Spectrum Sampling rate Lower frequency of expected pitch range Upper frequency of expected pitch range Estimates pitch from using Harmonic Product Spectrum (HPS) method. Signal Index of the first sample in signal for processing Index of the last sample in signal for processing Lower frequency of expected pitch range Upper frequency of expected pitch range FFT size Estimates pitch from using Harmonic Product Spectrum (HPS) method. Spectrum Sampling rate Lower frequency of expected pitch range Upper frequency of expected pitch range Estimates pitch from based on spectral peaks. Signal Index of the first sample in signal for processing Index of the last sample in signal for processing Lower frequency of expected pitch range Upper frequency of expected pitch range FFT size Estimates pitch from based on spectral peaks. Spectrum Sampling rate Lower frequency of expected pitch range Upper frequency of expected pitch range Estimates pitch from based on its cepstrum. Signal Index of the first sample in signal for processing Index of the last sample in signal for processing Lower frequency of expected pitch range Upper frequency of expected pitch range Size of cepstrum FFT size Provides methods for computing spectral features. Computes spectral centroid. Spectrum Center frequencies Computes spectral spread. Spectrum Center frequencies Computes spectral decrease. Spectrum Computes spectral flatness. Spectrum Amplitude threshold Computes spectral noiseness. Spectrum Center frequencies Lower frequency of noise Computes spectral rolloff frequency. Spectrum Center frequencies Rolloff percent Computes spectral crest. Spectrum Computes array of spectral contrasts in spectral bands. Spectrum Center frequencies Starting frequency Number of spectral bands Computes spectral contrast in a spectral band with index . Spectrum Center frequencies Spectral band index Starting frequency Computes Shannon entropy of a spectrum (spectrum is treated as p.d.f.) Spectrum Abstract class for adaptive filters. Constructs of given . Filter order Inits weights of adaptive filter. Weights (filter kernel) Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents LMF Adaptive filter (Least-Mean-Fourth algorithm). Constructs of given . Filter order Mu Leakage Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents LMS Adaptive filter (Least-Mean-Squares algorithm). Constructs of given . Filter order Mu Leakage Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents NLMF Adaptive filter (Normalized Least-Mean-Fourth algorithm + Epsilon). Constructs of given . Filter order Mu Epsilon Leakage Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents NLMS Adaptive filter (Normalized Least-Mean-Squares algorithm + Epsilon). Constructs of given . Filter order Mu Epsilon Leakage Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents Adaptive RLS filter (Recursive-Least-Squares algorithm). Lambda. Inverse corr matrix. Matrix of gain coefficients. Temporary matrices for calculations. Temporary matrices for calculations. Constructs of given . Order Lambda Value to initialize inverse corr matrix Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents Sign LMS Adaptive filter (Sign Least-Mean-Squares algorithm). Constructs of given . Filter order Mu Leakage Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents LMS Adaptive filter (Least-Mean-Squares) with variable steps. Constructs of given . Filter order Mu Leakage Processes one sample of input and desired signals and adapts filter coefficients. Sample of input signal Sample of desired signal Represents the chain of sequentially connected filters (double precision). List of filters in the chain. Constructs from collection of . Collection of online filters Constructs from collection of transfer functions (e.g., SOS sections). This constructor creates objects of under the hood. Collection of transfer functions Adds to the chain. Online filter Inserts at specified in the chain. Index of the filter in chain Online filter Removes filter at specified from the chain. Index of the filter in chain Processes one sample by the chain of filters. Input sample Resets all filters in the chain. Applies filters to entire and returns new filtered signal. Signal Filtering method Represents Finite Impulse Response (FIR) filter. Gets copy of the filter kernel (impulse response). Numerator part coefficients in filter's transfer function (non-recursive part in difference equations). Kernel length. Transfer function. Gets transfer function. Gets or sets the minimum kernel length for switching to OverlapSave algorithm in auto mode. Internal buffer for delay line. Current offset in delay line. Constructs from . FIR filter kernel Constructs from transfer function . Coefficients (used for filtering) will be cast to floats anyway, but filter will store the reference to TransferFunction object for FDA. Transfer function Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Processes all in loop. Samples Changes filter kernel online. New kernel Resets filter. Interface for offline filters (double precision). Applies filter to entire and returns new filtered signal. Signal Filtering method Represents Infinite Impulse Response (IIR) filter (double precision). Numerator part coefficients in filter's transfer function (non-recursive part in difference equations). Denominator part coefficients in filter's transfer function (recursive part in difference equations). Number of numerator coefficients. Number of denominator (feedback) coefficients. Transfer function. Gets transfer function. Gets or sets default length of truncated impulse response. Internal delay line (recursive part). Internal delay line (non-recursive part). Current offset in delay line (recursive part). Current offset in delay line (non-recursive part). Constructs from numerator and denominator . Numerator of transfer function Denominator of transfer function Constructs from transfer function . Transfer function Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Changes filter coefficients online (numerator / non-recursive part). New coefficients Changes filter coefficients online (denominator / recursive part). New coefficients Resets filter. Normalizes transfer function (divides all filter coefficients by the first coefficient of TF denominator). Interface for all signal processors that support online filtering (double precision). Processes one sample. Input sample Resets filter. Abstract class for Linear Time-Invariant (LTI) filters (double precision). Gets transfer function of LTI filter. Processes one sample. Input sample Resets LTI filter. Applies filter to entire and returns new filtered signal. Signal Filtering method Represents filter for processing data in interleaved stereo buffers. is wrapped around two separate filters: filter for signal in left channel and filter for signal in right channel. Filter for signal in left channel. Filter for signal in right channel. Internal flag for switching between left and right channels. Constructs from two separate filters. Filter for signal in left channel Filter for signal in right channel Processes one sample. Input sample Resets filters. Applies filter to entire and returns new filtered signal. Signal Filtering method is the special implementation of an LTI filter based on state vector (instead of delay lines). allows setting initial state (initial conditions for filter delays) and provides additional method for zero-phase filtering . Numerator part coefficients in filter's transfer function (non-recursive part in difference equations). Denominator part coefficients in filter's transfer function (recursive part in difference equations). State vector. Gets state vector. Transfer function. Gets transfer function. Constructs from numerator and denominator . Numerator of transfer function Denominator of transfer function Constructs from transfer function . Transfer function Initializes filter with initial conditions . Vector of initial conditions Processes one sample. Input sample Does zero-phase filtering (analog of filtfilt() in MATLAB/sciPy). Input signal Number of elements by which to extend at both ends before applying the filter. The default value is 3 * (max{len(numerator), len(denominator)} - 1). Changes filter coefficients online (numerator / non-recursive part). New coefficients Changes filter coefficients online (denominator / recursive part). New coefficients Changes filter coefficients online (from transfer function ). Transfer function Resets filter. Applies filter to entire and returns new filtered signal. Signal Filtering method Represents the chain of sequentially connected filters. List of filters in the chain. Constructs from collection of . Collection of online filters Constructs from collection of transfer functions (e.g., SOS sections). This constructor creates objects of under the hood. Collection of transfer functions Adds to the chain. Online filter Inserts at specified in the chain. Index of the filter in chain Online filter Removes filter at specified from the chain. Index of the filter in chain Processes one sample by the chain of filters. Input sample Resets all filters in the chain. Applies filters to entire and returns new filtered signal. Signal Filtering method Defines general filtering strategies. Filtering strategy is dynamically defined by NWaves library. Usually it's the processing of each signal sample in a loop. For longer FIR filter kernels it's the Overlap-Save algorithm. Filtering in time domain based on difference equations. Filtering in frequency domain based on OLA algorithm. Filtering in frequency domain based on OLS algorithm. Filtering strategy is defined fully in code by user. Represents Finite Impulse Response (FIR) filter. Gets copy of the filter kernel (impulse response). Numerator part coefficients in filter's transfer function (non-recursive part in difference equations). Kernel length. Transfer function. Gets transfer function. Gets or sets the minimum kernel length for switching to OverlapSave algorithm in auto mode. Internal buffer for delay line. Current offset in delay line. Constructs from . FIR filter kernel Constructs from 64-bit . NOTE. It will simply cast values to floats. If you need to preserve precision for filter design and analysis, use constructor . Constructs from transfer function . Coefficients (used for filtering) will be cast to floats anyway, but filter will store the reference to TransferFunction object for FDA. Transfer function Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Processes all in loop. Samples The most straightforward implementation of the difference equation: code the difference equation as it is (it's slower than ProcessAllSamples). Input signal Changes filter kernel online. New kernel Resets filter. Creates from sequential connection of two FIR filters and . First FIR filter Second FIR filter Creates from sequential connection of FIR and IIR . FIR filter IIR filter Creates from parallel connection of two FIR filters and . First FIR filter Second FIR filter Creates from parallel connection of FIR and IIR . FIR filter IIR filter Interface for offline filters. Applies filter to entire and returns new filtered signal. Signal Filtering method Provides extension methods for online filters. Filters data frame-wise. Online filter Input block of samples Block of filtered samples Number of samples to filter Input starting index Output starting index Filters entire by processing each signal sample in a loop. Online filter Input signal Filters entire by processing each signal sample in a loop. Online filter Input signal Calculates extra gain for filtering so that frequency response is normalized onto [0..1] range. Call this function if the filter could not be designed with proper frequency response gain due to numerical problems. Online filter FFT size (for evaluating frequency response) Filters entire with extra . Call this function if the filter could not be designed with proper frequency response gain due to numerical problems. Online filter Input signal Gain Processes one of a signal with extra . Call this function if the filter could not be designed with proper frequency response gain due to numerical problems. Online filter Input sample Gain Represents Infinite Impulse Response (IIR) filter. Numerator part coefficients in filter's transfer function (non-recursive part in difference equations). Denominator part coefficients in filter's transfer function (recursive part in difference equations). Number of numerator coefficients. Number of denominator (feedback) coefficients. Transfer function. Gets transfer function. Gets or sets default length of truncated impulse response. Internal delay line (recursive part). Internal delay line (non-recursive part). Current offset in delay line (recursive part). Current offset in delay line (non-recursive part). Constructs from numerator and denominator . Numerator of transfer function Denominator of transfer function Constructs from numerator and denominator (double precision). NOTE. It will simply cast values to floats. If you need to preserve precision for filter design and analysis, use constructor . Numerator of transfer function Denominator of transfer function Constructs from transfer function . Transfer function Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample The most straightforward implementation of the difference equation: code the difference equation as it is. Input signal Changes filter coefficients online (numerator / non-recursive part). New coefficients Changes filter coefficients online (denominator / recursive part). New coefficients Changes filter coefficients online (from transfer function ). Transfer function Resets filter. Normalizes transfer function (divides all filter coefficients by the first coefficient of TF denominator). Creates from sequential connection of IIR and any LTI . IIR filter LTI filter Creates from parallel connection of IIR and any LTI . IIR filter LTI filter Interface for all signal processors that support online filtering. Processes one sample. Input sample Resets filter. Abstract class for Linear Time-Invariant (LTI) filters. Gets transfer function of LTI filter. Applies LTI filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Resets LTI filter. Abstract class for all filters working by the STFT overlap-add scheme: short-time frame analysis short-time frame processing short-time frame synthesis (overlap-add) STFT hop length. Size of FFT for analysis and synthesis. ISTFT normalization gain. Size of frame overlap. Internal FFT transformer. Window coefficients. Delay line. Offset in the input delay line. Offset in the output buffer. Constructs . Hop size (hop length, number of samples) FFT size Processes one sample. Input sample Processes one frame (FFT block). Processes one spectrum at each Overlap-Add STFT step. Real parts of input spectrum Imaginary parts of input spectrum Real parts of output spectrum Imaginary parts of output spectrum Resets filter internals. Applies filter to entire and returns new filtered signal. Signal Filtering method State-space representation of filter. Gets or sets state matrix. Gets or sets input-to-state matrix. Gets or sets state-to-output matrix. Gets or sets feedthrough matrix. Represents filter for processing data in interleaved stereo buffers. is wrapped around two separate filters: filter for signal in left channel and filter for signal in right channel. Filter for signal in left channel. Filter for signal in right channel. Internal flag for switching between left and right channels. Constructs from two separate filters. Filter for signal in left channel Filter for signal in right channel Processes one sample. Input sample Resets filters. Applies filter to entire and returns new filtered signal. Signal Filtering method Represents the transfer function of an LTI filter. Gets numerator of transfer function. Gets denominator of transfer function. Gets or sets max number of iterations for calculating zeros/poles (roots of polynomials). By default, 25000. Zeros. Poles. Gets zeros ('z' in 'zpk' notation). Gets poles ('p' in 'zpk' notation). Gets gain ('k' in 'zpk' notation). Constructs from and . Numerator of transfer function Denominator of transfer function Constructs from , and . Zeros Poles Gain Constructs from , and . Zeros Poles Gain Constructs from representation. State space representation Gets state-space representation. Gets initial state for that corresponds to the steady state of the step response. Evaluates impulse response of given . In case of FIR filters method returns full copy of numerator. Length of the impulse response Evaluates frequency response of given . Length of the frequency response Evaluates group delay in array of given . Length of group delay array Evaluates phase delay in array of given . Length of phase delay array Normalizes frequency response at given frequency (normalizes numerator to map frequency response onto [0, 1]) Frequency Normalizes numerator and denominator (divides them by the first coefficient of denominator). Converts zeros (or poles) to numerator (or denominator) of transfer function. Zeros (or poles) Converts zeros (or poles) to numerator (or denominator) of transfer function. Zeros (or poles) Converts zeros (or poles) to numerator (or denominator) of transfer function. Zeros (poles) are given in the form double arrays of real and imaginary parts. Real parts of complex zeros (poles) Imaginary parts of complex zeros (poles) Converts numerator (or denominator) of transfer function to zeros (or poles). Numerator or denominator (polynomial) Max number of iterations for calculating zeros/poles (roots of polynomials). By default, 25000. Creates transfer function from sequential connection of and . First transfer function Second transfer function Creates transfer function from parallel connection of and . First transfer function Second transfer function Loads numerator and denominator of transfer function from csv stream. Input stream Delimiter Serializes numerator and denominator of transfer function to csv stream. Output stream Delimiter is the special implementation of an LTI filter based on state vector (instead of delay lines). allows setting initial state (initial conditions for filter delays) and provides additional method for zero-phase filtering . Numerator part coefficients in filter's transfer function (non-recursive part in difference equations). Denominator part coefficients in filter's transfer function (recursive part in difference equations). State vector. Gets state vector. Transfer function. Gets transfer function. Constructs from numerator and denominator . Numerator of transfer function Denominator of transfer function Constructs from numerator and denominator (double precision). NOTE. It will simply cast values to floats. If you need to preserve precision for filter design and analysis, use constructor . Numerator of transfer function Denominator of transfer function Constructs from transfer function . Transfer function Initializes filter with initial conditions . Vector of initial conditions Initializes filter with initial conditions . Vector of initial conditions Processes one sample. Input sample Does zero-phase filtering (analog of filtfilt() in MATLAB/sciPy). Input signal Number of elements by which to extend at both ends before applying the filter. The default value is 3 * (max{len(numerator), len(denominator)} - 1). Changes filter coefficients online (numerator / non-recursive part). New coefficients Changes filter coefficients online (denominator / recursive part). New coefficients Changes filter coefficients online (from transfer function ). Transfer function Resets filter. Applies filter to entire and returns new filtered signal. Signal Filtering method Represents bandpass Bessel filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Represents bandstop Bessel filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Represents highpass Bessel filter. Gets cutoff frequency. Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Represents lowpass Bessel filter. Gets cutoff frequency. Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Bessel filter prototype. Gets -th coefficient of -th order Bessel polynomial. k n Evaluates analog poles of Bessel filter of given . Filter order Represents BiQuad allpass filter. Gets center frequency. Gets Q factor. Constructs . Normalized center frequency in range [0..0.5] Q factor Sets filter coefficients. Normalized center frequency in range [0..0.5] Q factor Changes filter coefficients online (preserving the state of the filter). Normalized center frequency in range [0..0.5] Q factor Represents BiQuad bandpass filter. Gets center frequency. Gets Q factor. Constructs . Normalized center frequency in range [0..0.5] Q factor Sets filter coefficients. Normalized center frequency in range [0..0.5] Q factor Changes filter coefficients online (preserving the state of the filter). Normalized center frequency in range [0..0.5] Q factor Represents BiQuad IIR filter. Constructs . Constructs from filter coefficients (numerator {B0, B1, B2} and denominator {A0, A1, A2}). B0 B1 B2 A0 A1 A2 Processes one sample. Input sample Resets filter. Changes filter coefficients online (preserving the state of the filter). B0 B1 B2 A0 A1 A2 Represents BiQuad highpass filter. Gets cutoff frequency. Gets Q factor. Constructs . Normalized cutoff frequency in range [0..0.5] Q factor Sets filter coefficients. Normalized cutoff frequency in range [0..0.5] Q factor Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Q factor Represents BiQuad high-shelving filter. Gets shelf midpoint frequency. Gets Q factor. Gets gain (in dB). Constructs . Normalized shelf midpoint frequency in range [0..0.5] Q factor Gain (in dB) Sets filter coefficients. Normalized shelf midpoint frequency in range [0..0.5] Q factor Gain (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized shelf midpoint frequency in range [0..0.5] Q factor Gain (in dB) Represents BiQuad lowpass filter. Gets cutoff frequency. Gets Q factor. Constructs . Normalized cutoff frequency in range [0..0.5] Q factor Sets filter coefficients. Normalized cutoff frequency in range [0..0.5] Q factor Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Q factor Represents BiQuad low-shelving filter. Gets shelf midpoint frequency. Gets Q factor. Gets gain (in dB). Constructs . Normalized shelf midpoint frequency in range [0..0.5] Q factor Gain (in dB) Sets filter coefficients. Normalized shelf midpoint frequency in range [0..0.5] Q factor Gain (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized shelf midpoint frequency in range [0..0.5] Q factor Gain (in dB) Represents BiQuad notch filter. Gets center frequency. Gets Q factor. Constructs . Normalized center frequency in range [0..0.5] Q factor Sets filter coefficients. Normalized center frequency in range [0..0.5] Q factor Changes filter coefficients online (preserving the state of the filter). Normalized center frequency in range [0..0.5] Q factor Represents BiQuad peaking EQ filter. Gets center frequency. Gets Q factor. Gets gain (in dB). Constructs . Normalized center frequency in range [0..0.5] Q factor Gain (in dB) Sets filter coefficients. Normalized center frequency in range [0..0.5] Q factor Gain (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized center frequency in range [0..0.5] Q factor Gain (in dB) Represents bandpass Butterworth filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Represents bandstop Butterworth filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Represents highpass Butterworth filter. Gets cutoff frequency. Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Represents lowpass Butterworth filter. Gets cutoff frequency. Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Butterworth filter prototype. Evaluates analog poles of Butterworth filter of given . Filter order Represents bandpass Chebyshev-II filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Ripple (in dB) Represents bandstop Chebyshev-II filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Ripple (in dB) Represents highpass Chebyshev-II filter. Gets cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Ripple (in dB) Represents lowpass Chebyshev-II filter. Gets cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Ripple (in dB) Chebyshev-II filter prototype. Evaluates analog poles of Chebyshev-II filter of given . Filter order Ripple (in dB) Evaluates analog zeros of Chebyshev-II filter of given . Filter order Represents bandpass Chebyshev-I filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Ripple (in dB) Represents bandstop Chebyshev-I filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Ripple (in dB) Represents highpass Chebyshev-I filter. Gets cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Ripple (in dB) Represents lowpass Chebyshev-I filter. Gets cutoff frequency. Gets ripple (in dB). Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Ripple (in dB) Chebyshev-I filter prototype. Evaluates analog poles of Chebyshev-I filter of given . Filter order Ripple (in dB) Represents feedback comb filter: y[n] = b0 * x[n] - am * y[n - m] Delay (m). Constructs . Delay Coefficient b0 Coefficient am Processes one sample. Input sample Applies filter to entire and returns new filtered signal. Signal Filtering method Changes coefficients (preserving the state of the filter). Coefficient b0 Coefficient am Represents feedforward comb filter: y[n] = b0 * x[n] + bm * x[n - m] Delay (m). Constructs . Delay Coefficient b0 Coefficient bm Generates filter kernel. Delay Coefficient b0 Coefficient bm Normalize freq response to unit gain Processes one sample. Input sample Applies filter to entire and returns new filtered signal. Signal Filtering method Changes coefficients (preserving the state of the filter). Coefficient b0 Coefficient bm Represents DC removal IIR filter. R parameter. Constructs . R coefficient (usually in [0.9, 1] range) Processes one sample. Input sample Resets filter. Represents de-emphasis IIR filter. Constructs . De-emphasis coefficient Normalize freq response to unit gain Represents bandpass elliptic filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets passband ripple (in dB). Gets stopband ripple (in dB). Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Passband ripple (in dB) Stopband ripple (in dB) Represents bandstop elliptic filter. Gets low cutoff frequency. Gets high cutoff frequency. Gets passband ripple (in dB). Gets stopband ripple (in dB). Gets filter order. Constructs of given with given cutoff frequencies and . Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Generates transfer function. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Passband ripple (in dB) Stopband ripple (in dB) Represents highpass elliptic filter. Gets cutoff frequency. Gets passband ripple (in dB). Gets stopband ripple (in dB). Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Passband ripple (in dB) Stopband ripple (in dB) Represents lowpass elliptic filter. Gets cutoff frequency. Gets passband ripple (in dB). Gets stopband ripple (in dB). Gets filter order. Constructs of given with given cutoff . Normalized cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Generates transfer function. Normalized cutoff frequency in range [0..0.5] Filter order Passband ripple (in dB) Stopband ripple (in dB) Changes filter coefficients online (preserving the state of the filter). Normalized cutoff frequency in range [0..0.5] Passband ripple (in dB) Stopband ripple (in dB) Elliptic filter prototype. Evaluates analog poles of elliptic filter of given . Filter order Passband ripple (in dB) Stopband ripple (in dB) Evaluates analog zeros of elliptic filter of given . Filter order Passband ripple (in dB) Stopband ripple (in dB) Computes Landen sequence. K Number of iterations Computes sde. X Landen sequence Computes sne. X Landen sequence Computes inverse sne. X K Number of iterations Provides methods for filter design and analysis. Provides methods for filter design and analysis. Designs ideal lowpass fractional-delay FIR filter using sinc-window method. Filter order Normalized cutoff frequency in range [0..0.5] Fractional delay Window Designs ideal highpass fractional-delay FIR filter using sinc-window method. Filter order Normalized cutoff frequency in range [0..0.5] Fractional delay Window Designs ideal bandpass fractional-delay FIR filter using sinc-window method. Filter order Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Fractional delay Window Designs ideal bandstop fractional-delay FIR filter using sinc-window method. Filter order Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Fractional delay Window Designs ideal allpass fractional-delay FIR filter using sinc-window method. Filter order Fractional delay Window Normalizes frequency response at given frequency (normalizes kernel coefficients to map frequency response onto [0, 1]) Kernel Frequency Designs ideal lowpass FIR filter using sinc-window method. Filter order Normalized cutoff frequency in range [0..0.5] Window Designs ideal highpass FIR filter using sinc-window method. Filter order Normalized cutoff frequency in range [0..0.5] Window Designs ideal bandpass FIR filter using sinc-window method. Filter order Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Window Designs ideal bandstop FIR filter using sinc-window method. Filter order Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Window Designs FIR filter using frequency sampling method (works identical to firwin2 in sciPy and fir2 in MATLAB). Note. By default, the FFT size is auto-computed. If it is set explicitly, then (fftSize/2 + 1) must exceed the filter order. Note. Array of frequencies can be null. In this case the must be provided and size of gains array must be fftSize/2 + 1. Frequencies will be uniformly sampled on range [0..0.5]. Filter order Frequencies (frequency sampling points), each in range [0..0.5] Filter gains at the frequency sampling points FFT size Window Designs equiripple lowpass FIR filter using Remez (Parks-McClellan) algorithm. Filter order Passband edge frequency Stopband edge frequency Passband weight Stopband weight Designs equiripple highpass FIR filter using Remez (Parks-McClellan) algorithm. Filter order Stopband edge frequency Passband edge frequency Stopband weight Passband weight Designs equiripple bandpass FIR filter using Remez (Parks-McClellan) algorithm. Filter order Left stopband edge frequency Passband left edge frequency Passband right edge frequency Right stopband edge frequency Left stopband weight Passband weight Right stopband weight Designs equiripple bandstop FIR filter using Remez (Parks-McClellan) algorithm. Filter order Left passband edge frequency Stopband left edge frequency Stopband right edge frequency Right passband edge frequency Left passband weight Stopband weight Right passband weight Converts linear-phase lowpass filter to highpass filter with the same cutoff frequency. This method works only for odd-sized kernels. Lowpass filter kernel Converts linear-phase highpass filter to lowpass filter with the same cutoff frequency. This method works only for odd-sized kernels. Highpass filter kernel Converts linear-phase bandpass filter to bandstop filter with the same cutoff frequencies. This method works only for odd-sized kernels. Bandpass filter kernel Converts linear-phase bandstop filter to bandpass filter with the same cutoff frequencies. This method works only for odd-sized kernels. Bandstop filter kernel Designs IIR notch filter. Normalized center frequency in range [0..0.5] Q factor (characterizes notch filter -3dB bandwidth relative to its center frequency) Designs IIR peak filter. Normalized center frequency in range [0..0.5] Q factor (characterizes peak filter -3dB bandwidth relative to its center frequency) Designs IIR comb notch filter. Normalized center frequency in range [0..0.5] Q factor (characterizes notch filter -3dB bandwidth relative to its center frequency) Designs IIR comb peak filter. Normalized center frequency in range [0..0.5] Q factor (characterizes peak filter -3dB bandwidth relative to its center frequency) Designs lowpass pole filter. Normalized cutoff frequency in range [0..0.5] Analog prototype poles Analog prototype zeros Designs highpass pole filter. Normalized cutoff frequency in range [0..0.5] Analog prototype poles Analog prototype zeros Designs bandpass pole filter. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Analog prototype poles Analog prototype zeros Designs bandstop pole filter. Normalized low cutoff frequency in range [0..0.5] Normalized high cutoff frequency in range [0..0.5] Analog prototype poles Analog prototype zeros Converts second-order sections to transfer function (zeros-poles-gain). Array of SOS transfer functions Converts transfer function (zeros-poles-gain) to second-order sections. Transfer function Leaves only one of two conjugated numbers in the list of complex numbers. List of complex numbers Contains methods providing general shapes of filter banks: triangular rectangular FIR bandpass (close to trapezoidal, slightly overlapping) BiQuad bandpass ...and methods for obtaining the most widely used frequency bands: Herz bands Mel bands (HTK and Slaney) Bark bands (uniform and Slaney) Critical bands ERB filterbank Octaves (from MPEG-7) Chroma Generates triangular filterbank weights based on given . Assumed size of FFT Assumed sampling rate of a signal Array of frequency tuples (left, center, right) for each filter VTLN frequency warper Frequency scale mapper (e.g. herz-to-mel) used here only for proper weighting Generates rectangular filterbank weights based on given . Assumed size of FFT Assumed sampling rate of a signal Array of frequency tuples (left, center, right) for each filter VTLN frequency warper Frequency scale mapper (e.g. herz-to-mel) Generates FIR bandpass (close to trapezoidal) filterbank based on given . Assumed size of FFT Assumed sampling rate of a signal Array of frequency tuples (left, center, right) for each filter VTLN frequency warper Frequency scale mapper (e.g. herz-to-mel) Generates BiQuad bandpass overlapping filters based on given . Assumed size of FFT Assumed sampling rate of a signal Array of frequency tuples (left, center, right) for each filter Returns frequency tuples for uniformly spaced frequency bands on any scale. The function that converts Hz to other frequency scale The function that converts frequency from alternate scale back to Hz Number of filters Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Flag indicating that bands should overlap Returns frequency tuples for uniformly spaced frequency bands on Herz scale. Number of filters Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Flag indicating that bands should overlap Returns frequency tuples for uniformly spaced frequency bands on Mel scale. Number of mel filters to create Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Flag indicating that bands should overlap Returns frequency tuples for uniformly spaced frequency bands on Mel scale (according to M.Slaney's formula). Number of mel filters to create Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Flag indicating that bands should overlap Returns frequency tuples for uniformly spaced frequency bands on Bark scale (Traunmueller, 1990). Number of bark filters to create Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Flag indicating that bands should overlap Returns frequency tuples for uniformly spaced frequency bands on Bark scale (Wang, 1992). Number of bark filters to create Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Flag indicating that bands should overlap Returns frequency tuples for critical bands. Number of filters to create Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Returns frequency tuples for octave bands. Number of octave filters to create Assumed sampling rate of a signal Lower bound of the frequency range Upper bound of the frequency range Flag indicating that bands should overlap Generates chroma feature filter bank (slightly over-complicated to match librosa analog). Assumed size of FFT Assumed sampling rate Number of chroma features (12 pitch classes by default) Tuning deviation from A440 in fractions of a chroma bin If octaveWidth=0, centerOctave is ignored. Otherwise, it's the center of Gaussian window If octaveWidth=0, the shape is rectangular. Otherwise, it's the width of Gaussian window Norm: 0 - no normalization, 1 - apply L1-norm, 2 - apply L2-norm If baseC=true, the filter bank will start at 'C'. Otherwise, the filter bank will start at 'A'. Creates overlapping triangular mel filters (as suggested by Malcolm Slaney). Number of mel filters Assumed size of FFT Assumed sampling rate Lower bound of the frequency range Upper bound of the frequency range True if gain should be normalized; false if all filters should have same height 1.0 VTLN frequency warper Creates overlapping trapezoidal bark filters (as suggested by Malcolm Slaney). Number of bark filters Assumed size of FFT Assumed sampling rate Lower bound of the frequency range Upper bound of the frequency range Constant width of each band in Bark Creates overlapping ERB filters. Number of ERB filters Assumed size of FFT Assumed sampling rate Lower bound of the frequency range Upper bound of the frequency range True if gain should be normalized; false if all filters should have same height 1.0 Normalizes weights (so that energies in each band are approx. equal). Number of filters Array of frequency tuples (left, center, right) for each filter Filter bank Applies filters to spectrum and fills resulting filtered spectrum. Filter bank Spectrum Spectrum of filtered signal Applies filters to all spectra in given sequence. Filter bank Output spectra of filtered signal Applies filters to spectrum and then does Ln() on resulting spectrum. Filter bank Spectrum Spectrum of filtered signal Log-floor (Threshold for log-operation) Applies filters to spectrum and then does Log10() on resulting spectrum. Filter bank Spectrum Spectrum of filtered signal Log-floor (Threshold for log-operation) Applies filters to spectrum and then does 10*Log10() on resulting spectrum (added to compare MFCC coefficients with librosa results). Filter bank Spectrum Spectrum of filtered signal Threshold for log-operation Applies filters to spectrum and then does Pow(x, power) on resulting spectrum. For example, in PLP: power=1/3 (cubic root). Filter bank Spectrum Spectrum of filtered signal Power Optimal equiripple filter designer based on Remez (Parks-McClellan) algorithm. Example:

var order = 57;
var freqs = new double[] { 0, 0.15, 0.17, 0.5 };
var response = new double[] { 1, 0 };
var weights = new double[] { 0.01, 0.1 };

var remez = new Remez(order, freqs, response, weights);

var kernel = remez.Design();


// We can monitor the following properties:

remez.Iterations
remez.ExtremalFrequencies
remez.InterpolatedResponse
remez.Error
Gets filter order. Gets number of actual iterations. Gets number of extremal frequencies (K = Order/2 + 2). Gets interpolated frequency response. Gets array of errors. Gets array of extremal frequencies. Tolerance (for computing denominators). Indices of extremal frequencies in the grid. Grid. Band edge frequencies. Desired frequency response on entire grid. Weights on entire grid. Points for interpolation. Gamma coefficients used in Lagrange interpolation. Precomputed cosines. Constructs filter designer. Order of filter Array of normalized frequencies Array of desired response values at given frequencies Array of weights at given frequencies Grid density Creates grid (uniform in each band). Array of desired response values Array of weights Grid density Uniform initialization of extremal frequencies. Designs optimal equiripple filter and returns the kernel of designed filter. Max number of iterations Updates gamma coefficients, interpolation points and delta. Reconstructs impulse response from interpolated frequency response. Computes gamma coefficient. Input value Barycentric Lagrange interpolation. Frequency Converts ripple (in dB) to passband weight. Ripple (in dB) Converts ripple (in dB) to stopband weight. Ripple (in dB) Estimates order of a low-pass filter. Passband edge frequency Stopband edge frequency Passband weight Stopband weight Estimates order of a filter with custom bands. Parameters are given in conventional format. For example: frequencies: { 0, 0.2, 0.22, 0.32, 0.33, 0.5 }
deltas: { 0.01, 0.1, 0.06 } Array of edge frequencies Array of weights Vocal Tract Length Normalization (VTLN) similar to Kaldi implementation. Lower frequency. Upper frequency. Lower frequency for VTLN. Upper frequency for VTLN. Constructs . Warping factor Lower frequency Upper frequency Lower frequency for VTLN Upper frequency for VTLN Warps . Represents Hilbert filter. Gets size of the filter. Constructs of given . Size of the filter Generates filter kernel of given . Kernel size Represents median filter. Gets the size of median filter. Delay line (circular buffer). Buffer filled with sorted values from delay line. Index of the current sample. Constructs of given . Size of the filter Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Resets filter. Provides alternate implementation of median filter. It is slightly faster than only for small filter sizes (not exceeding 5, approx.). In other cases should be preferred. Gets the size of median filter. Constructs of given . Size of the filter Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Resets filter. Provides non-recursive implementation of moving-average filter. Gets size of the filter. Constructs of given . Size of the filter Generates filter kernel of given . Kernel size Provides fast recursive implementation of moving average filter: y[n] = x[n] / N - x[n - N] / N + y[n - 1]
i.e.
b = [1/N, 0, 0, 0, 0, ... , 0, -1/N]
a = [1, -1]
Gets size of the filter. Delay line. Constructs of given . Size of the filter Generates numerator of transfer function. Numerator size Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Resets filter. Represents one-pole highpass filter. Gets cutoff frequency. Constructs with given cutoff . Cutoff frequency Sets filter coefficients based on given cutoff . Cutoff frequency Changes filter coefficients (preserving the state of the filter). Cutoff frequency Represents one-pole lowpass filter. Gets cutoff frequency. Constructs with given cutoff . Cutoff frequency Sets filter coefficients based on given cutoff . Cutoff frequency Changes filter coefficients (preserving the state of the filter). Cutoff frequency Represents one-pole IIR filter. Delay line. Constructs . Constructs from filter coefficients. Numerator coefficient Pole Processes one sample. Input sample Resets filter. Represents the system of polyphase filters. Gets polyphase filters with transfer function E(z^k). Example:

h = [1, 2, 3, 4, 3, 2, 1], k = 3

e0 = [1, 0, 0, 4, 0, 0, 1]
e1 = [0, 2, 0, 0, 3, 0, 0]
e2 = [0, 0, 3, 0, 0, 2, 0]
Gets polyphase filters with transfer function E(z) used for multi-rate processing. h = [1, 2, 3, 4, 3, 2, 1], k = 3

e0 = [1, 4, 1]
e1 = [2, 3, 0]
e2 = [3, 2, 0]
Constructs with filters from filter . Filter kernel Number of polyphase filters Polyphase system type (1 or 2) Does polyphase decimation (for type-I systems). Input signal Does polyphase interpolation (for type-II systems). Input signal Processes one sample. Input sample Resets polyphase filters. Processes entire and returns new filtered signal. Input signal Filtering method Represents pre-emphasis FIR filter. Delay line (consisting of one sample). Constructs . Pre-emphasis coefficient Normalize freq response to unit gain Processes one sample. Input sample Applies filter to entire and returns new filtered signal. Signal Filtering method Resets filter. Represents RASTA filter (used for robust speech processing). Constructs . Pole Represents Savitzky-Golay filter. Gets size of the filter. Constructs of given . Size of the filter (must be odd number in range [5..31]) Derivative (must be 0, 1 or 2) Generates filter kernel. Kernels are precomputed. Size of the filter (must be odd number in range [5..31]) Derivative (must be 0, 1 or 2) Represents N-th order Thiran allpass interpolation filter for Delta delay samples. Example:
N = 13
Delta = 13 + 0.4
Constructs of given . Filter order Delta (fractional delay) Generates transfer function. Filter order Delta (fractional delay) Computes -th coefficient of denominator of transfer function. k n Delta Represents Wiener filter. Implementation is identical to scipy.signal.wiener(). Size of the Wiener filter. Estimated noise power. Constructs of given . Size of the Wiener filter Estimated noise power Applies filter to entire and returns new filtered signal. Signal Filtering method Processes one sample. Input sample Resets filter. Represents fast (FFT) convolver for complex-valued inputs/outputs. does not participate in heavy calculations, so it does not contain internal buffers. Memory is allocated for each operation ad-hoc. Does fast convolution of with via FFT. Does fast cross-correlation between and via FFT. Does fast deconvolution of from via polynomial division and FFT. Represents fast (FFT) convolver. FFT size. Internal FFT transformer. Constructs . Allocates necessary memory according to . If is not set then the memory will be allocated during the first call of Convolve() method based on input signals. Prepares all necessary arrays for calculations. Does fast convolution of with via FFT. Returns signal of length: signal.Length + kernel.Length - 1. Does fast convolution of with via FFT (maximally in-place). The result is stored in array. This version is best suited for block processing when memory needs to be reused. Input arrays must have size equal to the size of FFT. FFT size MUST be set explicitly and properly in constructor! Does fast cross-correlation between and via FFT. Does fast cross-correlation between and via FFT (maximally in-place). The result is stored in array. This version is best suited for block processing when memory needs to be reused. Input arrays must have size equal to the size of FFT. FFT size MUST be set explicitly and properly in constructor! Represents signal processor that implements Overlap-Add algorithm of block convolution. Filter kernel. FFT size. Internal FFT transformer. Offset in the input delay line. Offset in the delay line. Gets hop length: FFT size - kernel size + 1. Constructs with given and . Constructs with given and . Constructs with given FIR kernel and . Changes coefficients online. Processes one sample. Input sample Processes one frame (block). Processes entire and returns new filtered signal. Input signal Filtering method Resets Overlap-Add convolver. Represents signal processor that implements Overlap-Add algorithm of block convolution. Filter kernel. FFT size. Internal FFT transformer. Offset in the input delay line. Offset in the delay line. Gets hop length: FFT size - kernel size + 1. Constructs with given and . Constructs with given FIR kernel and . Changes coefficients online. Processes one sample. Input sample Processes one frame (block). Processes entire and returns new filtered signal. Input signal Filtering method Resets Overlap-Add convolver. Represents signal processor that implements Overlap-Save algorithm of block convolution. Filter kernel. FFT size. Internal FFT transformer. Offset in the input delay line. Offset in the delay line. Gets hop length: FFT size - kernel size + 1. Constructs with given and . Constructs with given and . Constructs with given FIR kernel and . Changes coefficients online. Processes one sample. Input sample Process one frame (block). Processes entire and returns new filtered signal. Input signal Filtering method Resets Overlap-Save convolver. Represents signal processor that implements Overlap-Save algorithm of block convolution. Filter kernel. FFT size. Internal FFT transformer. Offset in the input delay line. Offset in the delay line. Gets hop length: FFT size - kernel size + 1. Constructs with given and . Constructs with given FIR kernel and . Changes coefficients online. Processes one sample. Input sample Processes one frame (block). Processes entire and returns new filtered signal. Input signal Filtering method Resets Overlap-Save convolver. Defines types (modes) of dynamics processors. Smaller ratios, like 1:1, 2:1. Bigger ratios, like 5:1, 10:1. Smaller ratios, like 1:1, 2:1. Very high ratios, like 5:1. Represents dynamics processor: limiter or compressor or expander or noise gate. Dynamics processor mode. Envelope follower. Sampling rate. Min threshold for dB amplitude. Attack/Release time coefficient. Gets or sets compression/expansion threshold. Gets or sets compression/expansion ratio. Gets or sets makeup gain. Gets or sets attack time. Gets or sets release time. Constructs in given . Type (mode) of dynamics processor Sampling rate Compression/expansion threshold Compression/expansion ratio Makeup gain Attack time (in seconds) Release time (in seconds) Min threshold for dB amplitude Processes one sample. Input sample Resets dynamics processor. Processes entire and returns new signal (dynamics). Input signal Filtering method Represents envelope follower (envelope detector). Gets or sets attack time (in seconds). Gets or sets release time (in seconds). Sampling rate. Current envelope sample. Attack coefficient. Release coefficient. Constructs . Sampling rate Attack time (in seconds) Release time (in seconds) Processes one sample. Input sample Resets envelope follower. Processes entire and returns new signal (envelope). Input signal Filtering method Reconstructs signal from a power (or magnitude) spectrogram using Griffin-Lim iterative algorithm. STFT transformer. Magnitude part of the spectrogram. Get or sets magnitude gain factor. Constructs for iterative signal reconstruction from . Spectrogram (list of spectra) Window size fro STFT Hop size for STFT Window for STFT Power (2 - Power spectra, otherwise - Magnitude spectra) Constructs for iterative signal reconstruction from . Spectrogram (list of spectra) STFT transformer Power (2 - Power spectra, otherwise - Magnitude spectra) Does one iteration of reconstruction and returns reconstructed signal at current step. Signal reconstructed at previous iteration Reconstructs signal from spectrogram iteratively. Number of iterations in Griffin-Lim algorithm Represents harmonic/percussive separator based on median filtering. Internal STFT transformer. Masking function. Median filter along time axis. Median filter along frequency axis. Constructs . FFT size Hop length (number of samples) Size of median filter along time axis Size of median filter along frequency axis Masking mode Evaluates harmonic and percussive mag-phase spectrograms from given . Both spectrogram objects share the same phase array. Extracts harmonic and percussive signals from given . Masking modes for HPS algorithm. Provides various modulation methods: ring amplitude frequency phase Does ring modulation (RM) and returns RM signal. Carrier signal Modulator signal Does amplitude modulation (AM) and returns AM signal. Carrier signal Modulator frequency Modulation index (depth) Does frequency modulation (FM) and returns FM signal. Baseband signal Carrier amplitude Carrier frequency Frequency deviation Does sinusoidal frequency modulation (FM) and returns sinusoidal FM signal. Carrier signal frequency Carrier signal amplitude Modulator frequency Modulation index (depth) Length of FM signal Sampling rate Does linear frequency modulation (FM) and returns FM signal. Carrier signal frequency Carrier signal amplitude Modulation index (depth) Length of FM signal Sampling rate Does phase modulation (PM) and returns PM signal. Baseband signal Carrier amplitude Carrier frequency Frequency deviation Does simple amplitude demodulation of based on Hilbert transform. Does simple frequency demodulation pf based on Hilbert transform. Provides methods for various DSP/audio operations: convolution cross-correlation block convolution deconvolution resampling time-stretching rectification envelope detection spectral subtraction normalization (peak / RMS) periodogram (Welch / Lomb-Scargle) Does fast convolution of with via FFT. Does fast convolution of with via FFT. Does fast convolution of with via FFT. Does fast cross-correlation between and via FFT. Does fast cross-correlation between and via FFT. Does block convolution of with (using either Overlap-Add or Overlap-Save algorithm). Signal Convolution kernel FFT size Block convolution method (OverlapAdd / OverlapSave) Deconvolves and . Signal Kernel Does interpolation of followed by lowpass filtering. Signal Interpolation factor (e.g. factor=2 if 8000 Hz -> 16000 Hz) Lowpass anti-aliasing filter Does decimation of preceded by lowpass filtering. Signal Decimation factor (e.g. factor=2 if 16000 Hz -> 8000 Hz) Lowpass anti-aliasing filter Does band-limited resampling of . Signal Desired sampling rate Lowpass anti-aliasing filter Order Does simple resampling of (as the combination of interpolation and decimation). Input signal Interpolation factor Decimation factor Lowpass anti-aliasing filter Does time stretching of with parameters set by user. Signal Stretch factor (ratio) Window size (for vocoders - FFT size) Hop length Algorithm for TSM Does time stretching of with auto-derived parameters. Signal Stretch factor (ratio) Algorithm for TSM Extracts the envelope of . Signal Attack time (in seconds) Release time (in seconds) Full-rectifies . Half-rectifies . De-noises using spectral subtraction. Subtracts from . Signal Noise signal FFT size Hop size (number of samples) Normalizes peak level. Samples Peak level in decibels (dbFS), e.g. -1dB, -3dB, etc. Normalizes peak level. Signal Peak level in decibels (dBFS), e.g. -1dB, -3dB, etc. Changes peak level relatively to input (in-place). Samples Peak change in decibels, e.g. -6dB - decrease peak level twice Changes peak level relatively to input . Signal Peak change in decibels, e.g. -6dB - decrease peak level twice Normalizes RMS. Samples RMS in decibels (dBFS), e.g. -6dB, -18dB, -26dB, etc. Normalizes RMS. Signal RMS in decibels (dBFS), e.g. -6dB, -18dB, -26dB, etc. Changes RMS relatively to input . Samples RMS change in decibels, e.g. -6dB - decrease RMS twice Changes RMS relatively to input . Signal RMS change in decibels, e.g. -6dB - decrease RMS twice Computes periodogram using Welch's method. If =0 then power spectrum is evaluated, otherwise power spectral density is evaluated. Signal Window size (number of samples) Hop size (number of samples) Windowing function FFT size If sampling rate=0 then power spectrum is evaluated, otherwise power spectral density is evaluated Computes the Lomb-Scargle periodogram. Sample times Signal values at sample times Angular frequencies for output periodogram Subtract mean from values before periodogram evaluation Normalize periodogram by the residuals of the data around a constant reference model(at zero) Represents signal resampler (sampling rate converter). Gets or sets the order of lowpass anti-aliasing FIR filter that will be created automatically if the filter is not specified explicitly. By default, 101. Does interpolation of followed by lowpass filtering. Signal Interpolation factor (e.g. factor=2 if 8000 Hz -> 16000 Hz) Lowpass anti-aliasing filter Does decimation of preceded by lowpass filtering. Signal Decimation factor (e.g. factor=2 if 16000 Hz -> 8000 Hz) Lowpass anti-aliasing filter Does band-limited resampling of . Signal Desired sampling rate Lowpass anti-aliasing filter Order Does simple resampling of (as the combination of interpolation and decimation). Input signal Interpolation factor Decimation factor Lowpass anti-aliasing filter Represents spectral subtraction filter. Gets or sets spectral floor (beta coefficient). Gets or sets min threshold for subtraction factor (alpha). Gets or sets max threshold for subtraction factor (alpha). Gets or sets min SNR value (in dB). Gets or sets max SNR value (in dB). Noise estimate. Constructs . Array of noise samples FFT size Hop length (number of samples) Constructs . Noise signal FFT size Hop length (number of samples) Processes one spectrum at each STFT step. Real parts of input spectrum Imaginary parts of input spectrum Real parts of output spectrum Imaginary parts of output spectrum Estimates power spectrum of . Array of noise samples Index of the first sample in array for processing Index of the last sample in array for processing Estimates power spectrum of signal. Noise signal Index of the first sample in signal Index of the last sample in signal Represents TSM processor based on Paul stretch algorithm. Randomizer for phases. Constructs . Stretch ratio Hop length at analysis stage FFT size Processes spectrum at each STFT step: simply randomizes phases. Resets TSM processor. Represents Phase Vocoder with identity phase locking [Puckette]. Array of spectrum magnitudes (at current step). Array of spectrum phases (at current step). Array of phase deltas. Array of peak positions (indices). Constructs . Stretch ratio Hop length at analysis stage FFT size Processes spectrum with phase-locking at each STFT step. Represents Phase Vocoder. Hop size at analysis stage (STFT decomposition). Hop size at synthesis stage (STFT merging). Size of FFT for analysis and synthesis. Stretch ratio. Internal FFT transformer. Window coefficients. ISTFT normalization gain Linearly spaced frequencies. Internal buffer for real parts of analyzed block. Internal buffer for imaginary parts of analyzed block. Array of phases computed at previous step. Array of new synthesized phases. Constructs . Stretch ratio Hop length at analysis stage FFT size Processes entire and returns new time-stretched signal. Processes one spectrum at each STFT step. This routine is different for different PV-based techniques. Resets phase vocoder. Defines time scale modification algorithms. Phase vocoder. Phase vocoder with phase-locking. Waveform similarity-based Synchrnoized Overlap-Add. Paul stretch. Represents TSM processor based on Waveform-Synchronized Overlap-Add (WSOLA) technique. Stretch ratio. Window size. Hop size at analysis stage (STFT decomposition). Hop size at synthesis stage (STFT merging). Maximum length of the fragment for search of the most similar waveform. True if parameters were set by user (not by default). Internal convolver (will be used for evaluating auto-correlation if the window size is too big). Cross-correlation signal. Constructs TSM processor. Stretch ratio Window size Hop size at analysis stage Max delta in WSOLA algorithm Constructs TSM processor and auto-derives WSOLA parameters for given ratio. Stretch ratio Prepares the internal convolver (for large window sizes). Processes entire and returns new time-stretched signal. Finds position of the best waveform similarity. Current window Previous window Max delta Represents wave shaper. Wave shaper is a filter that maps an input signal to the output signal by applying arbitrary mathematical function (shaping function) to the input signal. Constructs using . Wave shaping function Processes one sample. Input sample Resets wave shaper. Processes entire and returns new wave-shaped signal. Input signal Filtering method Represents ADSR envelope builder. Parameters that can be set in method : "attack", "a" (default: 0.2) "decay", "d" (default: 0.2) "sustain", "s" (default: 0.2) "release", "r" (default: 0.2) "attackAmp", "amp" (default: 1.5) Gets ADSR state (attack, decay, sustain, release). One-pole filter coefficients One-pole filter coefficients Constructs from ADSR parameters (in the form of number of samples). Number of samples for attack stage Number of samples for decay stage Number of samples for sustain stage Number of samples for release stage Constructs from ADSR parameters (in the form of duration in seconds). Duration of attack stage (seconds) Duration of decay stage (seconds) Duration of sustain stage (seconds) Duration of release stage (seconds) Generates new sample. Resets sample generator. Sets the sampling rate of the signal to build. Sampling rate Represents Additive White Gaussian Noise (AWGN) builder. Generates additive white gaussian noise using Box-Muller transform. Parameters that can be set in method : "mean", "mu" (default: 0.0) "sigma", "stddev" (default: 1.0) Mean. Standard deviation. Constructs . Generates new sample. Resets sample generator. Interface for online signal generators (one-sample providers). Generates new sample. Resets sample generator. Interface for signal builders (offline signal generators). Gets the length of the signal to build. Builds new signal. Abstract class for all NWaves-style sample generators / signal builders. Delay of the signal to build (used only in Build() method). Number of times to repeat the signal (used only in Build() method). List of signals to be superimposed with the signal to build (only in Build() method). Resulting signal. Dictionary of setters for each parameter. Gets the sampling rate of the signal. Gets the length of the signal (number of samples). Gets the duration of the signal (in seconds). Gets brief descriptions (or simply names) of parameters. Assigns value to parameter . Parameter name Parameter value Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Builds new entire signal. Sets the number of samples of the signal to build. Number of samples Sets the duration of the signal to build. Duration (in seconds) Sets the sampling rate of the signal to build. Sampling rate Sets the delay of the signal to build. Signal delay Adds another one signal to superimpose with the signal to build. Signal for superimposing Sets the number of times to repeat the signal to build. Number of times for repeating Represents builder of chirp signals. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "start", "f0", "freq0" (default: 100.0 Hz) "end", "f1", "freq1" (default: 1000.0 Hz) Lower amplitude level. Upper amplitude level. Start frequency. End frequency. Constructs . Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Start frequency and end frequency must be greater than zero. Represents builder of cosinusoidal signals. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "frequency", "freq" (default: 100.0 Hz) "phase", "phi" (default: 0.0) Lower amplitude level. Upper amplitude level. Frequency of the sinusoid (in Hz). Initial phase of the sinusoid (in radians). Constructs . Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Frequency must be greater than zero. Represents fade in/out decorator of signal builders. Example: var sine = new SineBuilder(...);
var fadeSine = new FadeInOutBuilder(sine).In(0.05).Out(0.2); Signal builder to decorate. Number of samples in fade-in section. Number of samples in fade-out section. Index of the sample in fade-in section. Index of the sample in fade-out section. Index of current sample. Constructs around . Underlying signal builder Generates new sample. Resets sample generator. Sets duration of fade-in section in seconds. Sets duration of fade-out section in seconds. Starts fading out. Is signal started fading. Is signal finished fading. Represents builder of signals that uses Karplus-Strong algorithm. Parameters that can be set in method : "frequency", "freq", "f" (default: 100.0 Hz) "stretch", "s" (default: 1.0) "feedback", "a" (default: 1.0) Frequency (in Hz). Stretch factor (0, +INF]. Feedback coefficient [0, 1]. Constructs . Constructs from array of . Array of samples Generates wave table of length . Number of wave table samples Generates new sample. Resets sample generator. Sets the sampling rate of the signal to build. Sampling rate Previous sample. Randomizer. Represents builder of signals that uses the "Drum" variation of Karplus-Strong algorithm. Parameters that can be set in method : "frequency", "freq", "f" (default: 100.0 Hz) "stretch", "s" (default: 1.0) "feedback", "a" (default: 1.0) "probability", "prob" (default: 0.5) Constructs . Constructs from array of . Array of samples Generates new sample. Represents builder of signals that uses PadSynth algorithm. Parameters that can be set in method : "frequency", "freq", "f" (default: 440.0 Hz) "fftsize", "size" (default: 2048) "bandwidth", "bw" (default: 40) "bwscale", "scale" (default: 1.25) Frequency of the note. Amplitudes of harmonics. Bandwidth of the first harmonic. how much the bandwidth of the harmonic increase according to it's frequency Internal FFT transformer. FFT size. Internal buffer for real parts of spectrum. Internal buffer for imaginary parts of spectrum. Constructs . Sets frequency of the note. Frequency Sets FFT size. Must be power of 2. FFT size Sets bandwidth. Bandwidth Sets the bandwidth scale parameter. Bandwidth scale Sets amplitudes of harmonics. Array of amplitudes Generates wave table using PadSynth algorithm. Helper method used in PadSynth algorithm. Frequency of the note Bandwidth Sets the sampling rate of the signal to build. Sampling rate Represents Perlin noise builder (1D simplex noise). Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "scale", "octave" (default: 0.02) Lower amplitude level. Upper amplitude level. Scale. Table of permutations. Constructs . 1D simplex noise. Gradient. Improved interpolator. Linear interpolator. Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Upper amplitude must be greater than lower amplitude. Represents pink noise builder. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) Lower amplitude level. Upper amplitude level. Constructs . Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Upper amplitude must be greater than lower amplitude. Represents builder of periodic pulse waves. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "pulse", "width" (default: 0.05 seconds) "period", "t" (default: 0.1 seconds) Lower amplitude level. Upper amplitude level. Pulse duration (in seconds). Period of pulse wave (in seconds). Constructs . Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Period and pulse duration must be greater than zero. Period must be greater than pulse duration. Represents builder of straight line signals: y[n] = slope * n + intercept. Parameters that can be set in method : "slope", "k" (default: 0.0) "intercept", "b" (default: 0.0) Slope. Intercept. Constructs . Generates new sample. Resets sample generator. Represents red (Brownian) noise builder. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) Lower amplitude level. Upper amplitude level. Constructs . Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Upper amplitude must be greater than lower amplitude. Represents sawtooth wave builder. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "frequency", "freq" (default: 100.0 Hz) Lower amplitude level. Upper amplitude level. Frequency of the sawtooth wave. Constructs . Generates new sample. Resets sample generator. Sets the sampling rate of the signal to build. Sampling rate Generates signal by generating all its samples one-by-one. Frequency must be greater than zero. Upper amplitude must be greater than lower amplitude. Represents builder of signal Sinc(x). Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "frequency", "freq" (default: 100.0 Hz) Lower amplitude level. Upper amplitude level. Frequency of the sinc (Hz). Constructs . Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Frequency must be greater than zero. Represents builder of sinusoidal signals. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "frequency", "freq" (default: 100.0 Hz) "phase", "phi" (default: 0.0) Lower amplitude level. Upper amplitude level. Frequency of the sinusoid (in Hz). Initial phase of the sinusoid (in radians). Constructs . Generates new sample. Resets sample generator. Generates signal by generating all its samples one-by-one. Frequency must be greater than zero. Represents builder of square waves. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "frequency", "freq" (default: 100.0 Hz) Lower amplitude level. Upper amplitude level. Frequency of the square wave (Hz). Constructs . Generates new sample. Resets sample generator. Sets the sampling rate of the signal to build. Sampling rate Generates signal by generating all its samples one-by-one. Frequency must be greater than zero. Upper amplitude must be greater than lower amplitude. Represents builder of triangle waves. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) "frequency", "freq" (default: 100.0 Hz) Lower amplitude level. Upper amplitude level. Frequency of the triangle wave. Constructs . Generates new sample. Resets sample generator. Sets the sampling rate of the signal to build. Sampling rate Generates signal by generating all its samples one-by-one. Frequency must be greater than zero. Upper amplitude must be greater than lower amplitude. Represents builder of signals that uses a wave table. Parameters that can be set in method : "stride", "step", "delta" (default: 1) Wave table samples. Stride. Interpolate sample or take the nearest one in the wave table. True if the stride is not integer. Constructs from . Wave table samples Generates new sample (take or interpolate sample from the wave table). Resets sample generator. Current offset. Represents white noise builder. Parameters that can be set in method : "low", "lo", "min" (default: -1.0) "high", "hi", "max" (default: 1.0) Lower amplitude level. Upper amplitude level. Constructs . Generates new sample. Generates signal by generating all its samples one-by-one. Upper amplitude must be greater than lower amplitude. Base class for finite complex-valued discrete-time signals. Finite complex DT signal is stored as two arrays of data (real parts and imaginary parts) sampled at certain sampling rate. See also for extra functionality of complex DT signals. Gets sampling rate (number of samples per one second). Gets the real parts of complex-valued samples. Gets the imaginary parts of complex-valued samples. Gets the length of the signal. The most efficient constructor for initializing complex discrete signals. By default, it just wraps around arrays and (without copying). If a new memory should be allocated for signal data, set to true. Sampling rate of the signal Array of real parts of the complex-valued signal Array of imaginary parts of the complex-valued signal Set to true if new memory should be allocated for data Constructs complex signal from collections of and parts. Sampling rate of the signal Array of real parts of the complex-valued signal Array of imaginary parts of the complex-valued signal Constructs signal from collection of sampled at . Sampling rate Collection of complex-valued samples Constructs signal of given filled with specified values. Sampling rate Number of samples Value of each sample Value of each sample Constructs signal from collection of integer sampled at given . Sampling rate Collection of integer samples Each sample will be divided by this value Creates deep copy of the signal. Sample indexer. Works only with array of real parts of samples. Use it with caution. Creates the slice of the signal: var middle = signal[900, 1200]; Index of the first sample (inclusive) Index of the last sample (exclusive) Gets the magnitudes of complex-valued samples. Gets the power (squared magnitudes) of complex-valued samples. Gets the phases of complex-valued samples. Gets the unwrapped phases of complex-valued samples. Creates new signal by superimposing signals and . If sizes are different then the smaller signal is broadcast to fit the size of the larger signal. First signal Second signal Creates new signal by adding to signal . Signal Constant to add to each sample Creates new signal by subtracting from signal . Signal Constant to subtract from each sample Creates new signal by multiplying by (amplification/attenuation). Signal Amplification/attenuation coefficient Provides extension methods for working with complex discrete signals. Creates the delayed copy of by shifting it either to the right (positive , e.g. Delay(1000)) or to the left (negative , e.g. Delay(-1000)). Signal Delay (positive or negative number of delay samples) Superimposes signals and . If sizes are different then the smaller signal is broadcast to fit the size of the larger signal. First signal Second signal Concatenates and . First signal Second signal Creates the copy of repeated times. Signal Number of times to repeat Amplifies by in-place. Signal Amplification coefficient Attenuates by in-place. Signal Attenuation coefficient Creates new signal from first samples of . Signal Number of samples to copy Creates new signal from last samples of . Signal Number of samples to copy Creates new zero-padded complex discrete signal of from . Signal The length of a zero-padded signal. Performs the complex multiplication of and (with normalization by length). First signal Second signal Performs the complex division of and (with normalization by length). First signal Second signal Unwraps phases of complex-valued samples. Phases Jump size Yields complex numbers as type from samples. Complex discrete signal Base class for finite real-valued discrete-time signals. In general, any finite DT signal is simply an array of data sampled at certain sampling rate. See also for extra functionality of DT signals. Gets sampling rate (number of samples per one second). Gets real-valued array of samples. Gets the length of the signal. Gets the duration of the signal (in seconds). The most efficient constructor for initializing discrete signals. By default, it just wraps around (without copying). If a new memory should be allocated for signal data, set to true. Sampling rate Array of samples Set to true if new memory should be allocated for signal data Constructs signal from collection of sampled at . Sampling rate Collection of samples Constructs signal of given filled with specified values. Sampling rate Number of samples Value of each sample Constructs signal from collection of integer sampled at given . Sampling rate Collection of integer samples Each sample will be divided by this value Generates unit impulse of given sampled at given . Length of unit impulse Sampling rate Generates constant signal of given sampled at given . Constant value Length of constant signal Sampling rate Creates deep copy of the signal. Sample indexer. Sample index Creates the slice of the signal: var middle = signal[900, 1200]; Index of the first sample (inclusive) Index of the last sample (exclusive) Creates new signal by superimposing signals and . If sizes are different then the smaller signal is broadcast to fit the size of the larger signal. First signal Second signal Creates negated copy of signal . Signal Subtracts signal from signal . If sizes are different then the smaller signal is broadcast to fit the size of the larger signal. First signal Second signal Creates new signal by adding to signal . Signal Constant to add to each sample Creates new signal by subtracting from signal . Signal Constant to subtract from each sample Creates new signal by multiplying by (amplification/attenuation). Signal Amplification/attenuation coefficient Computes energy of a signal fragment. Index of the first sample (inclusive) Index of the last sample (exclusive) Computes energy of entire signal. Computes RMS of a signal fragment. Index of the first sample (inclusive) Index of the last sample (exclusive) Computes RMS of entire signal. Computes Zero-crossing rate of a signal fragment. Index of the first sample (inclusive) Index of the last sample (exclusive) Computes Zero-crossing rate of entire signal. Computes Shannon entropy of a signal fragment (from bins distributed uniformly between the minimum and maximum values of samples). Index of the first sample (inclusive) Index of the last sample (exclusive) Number of bins Computes Shannon entropy of entire signal (from bins distributed uniformly between the minimum and maximum values of samples). Number of bins Provides extension methods for working with objects. Creates the delayed copy of by shifting it either to the right (positive , e.g. Delay(1000)) or to the left (negative , e.g. Delay(-1000)). Signal Delay (positive or negative number of delay samples) Superimposes signals and . If sizes are different then the smaller signal is broadcast to fit the size of the larger signal. First signal Second signal Superimposes and multiple times at given . First signal Second signal Positions (indices) where to insert Subtracts from . If sizes are different then the smaller signal is broadcast to fit the size of the larger signal. First signal Second signal Concatenates and . First signal Second signal Creates the copy of repeated times. Signal Number of times to repeat Amplifies by in-place. Signal Amplification coefficient Attenuates by in-place. Signal Attenuation coefficient Reverses in-place. Creates new signal from first samples of . Signal Number of samples to copy Creates new signal from last samples of . Signal Number of samples to copy Full-rectifies in-place. Signal Half-rectifies in-place. Signal Normalizes by its max absolute value (to range [-1, 1]). Signal Bit depth Creates from . Imaginary parts will be filled with zeros. Real-valued signal Fades signal in and out linearly (in-place). Signal Fade-in duration Fade-out duration Fades signal in linearly (in-place). Signal Fade-in duration (in seconds) Fades signal out linearly (in-place). Signal Fade-out duration (in seconds) Crossfades linearly between signals and returns crossfaded signal of length equal to sum of signal lengths minus length of crossfade section. The length of crossfade section will be calculated based on the sampling rate of the first signal. First signal Second signal Crossfade duration (in seconds) Interface for complex-valued transforms. Gets transform size. Does direct transform. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized direct transform. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does inverse transform. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized inverse transform. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Interface for real-valued transforms. Gets transform size. Does direct transform. Input data Output data Does normalized direct transform. Input data Output data Does inverse transform. Input data Output data Does normalized inverse transform. Input data Output data Represents Cepstral Transform (CT): Direct Complex CT (complex cepstrum) Inverse Complex CT Real cepstrum Power cepstrum Phase cepstrum 1,2) and 3) are identical to MATLAB functions cceps, icceps and rceps, respectively. operates on real-valued data. Gets cepstrum size. FFT transformer. Logarithm base (E or 10). Internal array for real parts of spectrum. Internal array for imaginary parts of spectrum. Internal array for storing the unwrapped phase. Constructs cepstral transformer. If exceeds , FFT size will be recalculated as the nearest power of 2 to cepstrum size. Cepstrum size FFT size Logarithm base Evaluates complex cepstrum as: Real{IFFT(log(abs(FFT(x)) + unwrapped_phase))} Input data Complex cepstrum Normalize cepstrum by FFT size Circular delay (number of samples) added to Evaluates inverse complex cepstrum of (removing samples). Input data Inverse complex cepstrum Normalize result by FFT size Delay (usually, returned by function ) Evaluates real cepstrum as: real{IFFT(log(abs(FFT(x))))} Input data Real cesptrum Normalize cepstrum by FFT size Evaluates power cepstrum as: power_cepstrum = 4 * real_cepstrum ^ 2 Evaluates phase cepstrum as: phase_cepstrum = (complex_cepstrum - reversed_complex_cepstrum) ^ 2 Input data Phase cepstrum Normalize cepstrum by FFT size Does cepstral transform. It simply calls ignoring the delay parameter. Input data Output data Does normalized cepstral transform. It simply calls ignoring the delay parameter. Input data Output data Does inverse cepstral transform. It simply calls ignoring the delay parameter. Input data Output data Does normalized inverse cepstral transform. It simply calls ignoring the delay parameter. Input data Output data Represents Discrete Cosine Transform of Type-I. DCT-I precalculated cosine matrix. Gets size of DCT-I. Constructs of given and precalculates DCT matrices. Size of DCT-I Does DCT-I. Input data Output data Does normalized DCT-I. Input data Output data Does Inverse DCT-I. Input data Output data Does normalized Inverse DCT-I. Input data Output data Represents Discrete Cosine Transform of Type-II. DCT-II precalculated cosine matrix. IDCT-II precalculated cosine matrix. Gets size of DCT-II. Constructs of given and precalculates DCT matrices. Size of DCT-II Does DCT-II. Input data Output data Does normalized DCT-II. Input data Output data Does Inverse DCT-II. Input data Output data Does normalized Inverse DCT-II. Input data Output data Represents Discrete Cosine Transform of Type-III. DCT-III precalculated cosine matrix. IDCT-III precalculated cosine matrix. Gets size of DCT-III. Constructs of given and precalculates DCT matrices. Size of DCT-III Does DCT-III. Input data Output data Does normalized DCT-III. Input data Output data Does Inverse DCT-III. Input data Output data Does normalized Inverse DCT-III. Input data Output data Represents Discrete Cosine Transform of Type-IV. DCT-IV precalculated cosine matrix. Gets size of DCT-IV. Constructs of given and precalculates DCT matrices. Size of DCT-IV Does DCT-IV. Input data Output data Does normalized DCT-IV. Input data Output data Does Inverse DCT-IV. Input data Output data Does normalized Inverse DCT-IV. Input data Output data Represents Discrete Cosine Transform of Type-II. This FFT-based implementation of DCT-II is faster for bigger DCT sizes. Internal FFT transformer. Internal temporary buffer. Gets size of DCT-II. Constructs of given . Size of DCT-II Does DCT-II. Input data Output data Does normalized DCT-II. Input data Output data Does Inverse DCT-II. Input data Output data Does normalized Inverse DCT-II. Input data Output data Represents Discrete Cosine Transform of Type-III. This FFT-based implementation of DCT-III is faster for bigger DCT sizes. Internal fast DCT-II transformer. Gets size of DCT-III. Constructs of given . Size of DCT-III Does DCT-III. Input data Output data Does normalized DCT-III. Input data Output data Does Inverse DCT-III. Input data Output data Does normalized Inverse DCT-III. Input data Output data Represents Discrete Cosine Transform of Type-IV. This FFT-based implementation of DCT-IV is faster for bigger DCT sizes. Internal FFT transformer. Gets size of DCT-IV. Constructs of given . Size of DCT-IV Does DCT-IV. Input data Output data Does normalized DCT-IV. Input data Output data Does Inverse DCT-IV. Input data Output data Does normalized Inverse DCT-IV. Input data Output data Represents Modified Discrete Cosine Transform (MDCT). This FFT-based implementation of MDCT is faster for bigger DCT sizes. Constructs of given . Size of MDCT Separate interface for DCT transforms. Represents Modified Discrete Cosine Transform (MDCT). Internal DCT-IV transformer. Internal temporary buffer. Gets size of MDCT. Constructs of given . Size of MDCT Internal DCT transformer (by default, ) Does MDCT. Length of must be equal to 2*. Length of must be equal to . Input data Output data Does normalized MDCT. Length of must be equal to 2*. Length of must be equal to . Input data Output data Does Inverse MDCT. Length of must be equal to . Length of must be equal to 2*. Input data Output data Does normalized Inverse MDCT. Length of must be equal to . Length of must be equal to 2*. Input data Output data Represents Complex Fast Fourier Transform: Direct FFT Inverse FFT Magnitude spectrum Power spectrum Gets FFT size. Precomputed cosines. Precomputed sines. Intermediate buffer storing real parts of spectrum. Intermediate buffer storing imaginary parts of spectrum. Constructs FFT transformer with given . FFT size must be a power of two. FFT size Does Fast Fourier Transform in-place. Array of real parts Array of imaginary parts Does Inverse Fast Fourier Transform in-place. Array of real parts Array of imaginary parts Does normalized Inverse Fast Fourier Transform in-place. Array of real parts Array of imaginary parts Does Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does Inverse Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Inverse Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Computes magnitude spectrum from : spectrum = sqrt(re * re + im * im) Method fills array . It must have size at least fftSize/2+1. Array of samples Magnitude spectrum Normalize by FFT size or not Computes power spectrum from : spectrum = (re * re + im * im) Method fills array . It must have size at least fftSize/2+1. Array of samples Magnitude spectrum Normalize by FFT size or not Computes and returns magnitude spectrum from : spectrum = sqrt(re * re + im * im) Signal Normalize by FFT size or not Computes and returns power spectrum from : spectrum = (re * re + im * im) Signal Normalize by FFT size or not FFT shift in-place. Throws if array of has odd length. Represents Complex Fast Fourier Transform for 64-bit data: Direct FFT Inverse FFT Magnitude spectrum Power spectrum Gets FFT size. Precomputed cosines. Precomputed sines. Constructs FFT transformer with given . FFT size must be a power of two. FFT size Does Fast Fourier Transform in-place. Array of real parts Array of imaginary parts Does Inverse Fast Fourier Transform in-place. Array of real parts Array of imaginary parts Does normalized Inverse Fast Fourier Transform in-place. Array of real parts Array of imaginary parts Does Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does Inverse Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Inverse Fast Fourier Transform: complex (, ) -> complex(, ). Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Provides implementation of Goertzel algorithm. FFT size. Constructs . FFT size Computes -th component of a spectrum using Goertzel algorithm. Input array of samples Number of the frequency component Computes -th component of a spectrum using Goertzel algorithm. Input signal Number of the frequency component Represents Fast Hartley Transform. Gets size of Hartley transform. Internal FFT transformer. Internal array for imaginary parts. Constructs Hartley transformer. Transform must be a power of 2. Size of Hartley transform Does Fast Hartley Transform in-place. Input/output data Does inverse Hartley transform in-place. Input/output data Does normalized Inverse Fast Hartley transform in-place. Input/output data Does Fast Hartley Transform. Input data Output data Does normalized Fast Hartley Transform. Identical to . Input data Output data Does Inverse Fast Hartley Transform. Input data Output data Does normalized Inverse Fast Hartley Transform. Input data Output data Represents Fast Hilbert Transform. Gets size of Hilbert transform. Internal FFT transformer. Intermediate buffer for real parts. Intermediate buffer for imaginary parts. Constructs Hilbert transformer. Transform must be a power of 2. Computes complex analytic signal (real and imaginary parts) from . Input data Does Fast Hilbert Transform. Input data Output data Does normalized Fast Hilbert Transform. Input data Output data Does Inverse Fast Hilbert Transform. Input data Output data Does normalized Inverse Fast Hilbert Transform. Input data Output data Represents Fast Hilbert Transform (for 64-bit data). Gets size of Hilbert transform. Internal FFT transformer. Intermediate buffer for real parts. Intermediate buffer for imaginary parts. Constructs Hilbert transformer. Transform must be a power of 2. Computes complex analytic signal (real and imaginary parts) from . Input data Does Fast Hilbert Transform. Input data Output data Does normalized Fast Hilbert Transform. Input data Output data Does Inverse Fast Hilbert Transform. Input data Output data Does normalized Inverse Fast Hilbert Transform. Input data Output data Represents Fast Mellin Transform. Gets the size of input data for Mellin transform. Gets the size of Mellin transform. Beta coefficient (0.5 by default, which corresponds to Scale transform). Time points on linear scale. Time points on exponential scale. FFT transformer. Constructs Mellin transformer. Parameter must be a power of 2. Expected size of input data Size of output data Beta coefficient (0.5 by default, which corresponds to Scale transform) Does Fast Mellin Transform. Input array of samples Output array of real parts Output array of imaginary parts Does normalized Fast Mellin Transform. Input array of samples Output array of real parts Output array of imaginary parts Does Fast Mellin Transform. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Fast Mellin Transform. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Inverse Fast Mellin Transform is not implemented. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Inverse normalized Fast Mellin Transform is not implemented. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Represents Complex Fast Fourier Transform (for real-valued input): Direct FFT for real-valued input Inverse FFT with real-valued output Magnitude spectrum Power spectrum Gets FFT size. Half of FFT size (for calculations). Precomputed cosines. Precomputed sines. Precomputed coefficients. Precomputed coefficients. Precomputed coefficients. Precomputed coefficients. Constructs FFT transformer with given . FFT size must be a power of two. FFT size Does Fast Fourier Transform: real -> complex (, ). Input data (real) Output data (real parts) Output data (imaginary parts) Does Inverse Fast Fourier Transform: complex (, ) -> real . Input data (real parts) Input data (imaginary parts) Output data (real) Does normalized Inverse Fast Fourier Transform: complex (, ) -> real . Input data (real parts) Input data (imaginary parts) Output data (real) Does Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does Inverse Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Inverse Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Computes magnitude spectrum from : spectrum = sqrt(re * re + im * im) Method fills array . It must have size at least fftSize/2+1. Array of samples Magnitude spectrum Normalize by FFT size or not Computes power spectrum from : spectrum = sqrt(re * re + im * im) Method fills array . It must have size at least fftSize/2+1. Array of samples Magnitude spectrum Normalize by FFT size or not Computes and returns magnitude spectrum from : spectrum = sqrt(re * re + im * im) Signal Normalize by FFT size or not Computes and returns power spectrum from : spectrum = (re * re + im * im) Signal Normalize by FFT size or not FFT shift in-place. Throws if array of has odd length. Represents Complex Fast Fourier Transform (for real-valued input, 64-bit): Direct FFT for real-valued input Inverse FFT with real-valued output Magnitude spectrum Power spectrum Gets FFT size. Half of FFT size (for calculations). Precomputed cosines. Precomputed sines. Precomputed coefficients. Precomputed coefficients. Precomputed coefficients. Precomputed coefficients. Constructs FFT transformer with given . FFT size must be a power of two. FFT size Does Fast Fourier Transform: real -> complex (, ). Input data (real) Output data (real parts) Output data (imaginary parts) Does Inverse Fast Fourier Transform: complex (, ) -> real . Input data (real parts) Input data (imaginary parts) Output data (real) Does normalized Inverse Fast Fourier Transform: complex (, ) -> real . Input data (real parts) Input data (imaginary parts) Output data (real) Does Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does Inverse Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Does normalized Inverse Fast Fourier Transform: complex (, ) -> complex(, ). is ignored. Input data (real parts) Input data (imaginary parts) Output data (real parts) Output data (imaginary parts) Represents Short-Time Fourier Transform. Gets FFT size. Internal FFT transformer. Overlap size (number of samples). Window size (number of samples). Window type. Pre-computed samples of the window function. Constructs STFT transformer. Size of analysis window Hop (overlap) size Type of the window function to apply Size of FFT Does STFT of an . Returns list of computed spectra (real and imaginary parts) in time. Input data Does STFT of a . Returns list of computed spectra (real and imaginary parts) in time. Input signal Does Inverse STFT from list of spectra . List of spectra (real and imaginary parts) Perfect reconstruction mode Helper method for ISTFT in 'perfect reconstruction' mode. Summed window coefficients Computes spectrogram. The spectrogram is essentially a list of power spectra in time. Input data Normalize each spectrum Computes spectrogram. The spectrogram is essentially a list of power spectra in time. Input signal Normalize each spectrum Computes averaged periodogram (used, for example, in Welch method). This method is memory-efficient since it doesn't store all spectra in memory. Input data Computes spectrogram in the form of list of magnitudes and phases from . Input data Computes spectrogram in the form of list of magnitudes and phases from . Input signal Reconstructs samples from in the form of list of magnitudes and phases. Spectrogram in the form of list of magnitudes and phases Perfect reconstruction mode Represents spectrogram in the form of list of magnitudes and phases. Gets or sets list of magnitudes. Gets or sets list of phases. Represents Fast Wavelet Transform (FWT). Gets FWT size. The length of the mother wavelet. LP coefficients for decomposition. HP coefficients for decomposition. LP coefficients for reconstruction. HP coefficients for reconstruction. Temporary internal buffer. Constructs FWT transformer. FWT size Mother wavelet Does Fast Wavelet Transform (decomposition). Input data Output data Does normalized Fast Wavelet Transform (decomposition). Identical to . Input data Output data Does inverse Fast Wavelet Transform (reconstruction). Input data Output data Does normalized inverse Fast Wavelet Transform (reconstruction). Identical to . Input data Output data Does Fast Wavelet Transform (decomposition). Input data Output data FWT level Does inverse Fast Wavelet Transform (reconstruction). Input data Output data FWT level Evaluates maximum decomposition level from input . Input length Represents wavelet. Gets wavelet name. Gets wavelet length. Gets LP coefficients for decomposition. Gets HP coefficients for decomposition. Gets LP coefficients for reconstruction. Gets HP coefficients for reconstruction. Constructs wavelet from and number of . Wavelet family Number of taps Constructs wavelet from . Supported names: "haar", "db1".."db20", "sym2".."sym20", "coif1".."coif5". Wavelet name Constructs wavelet from wavelet coefficients (perhaps, calculated in external software). LP coefficients for decomposition HP coefficients for decomposition LP coefficients for reconstruction HP coefficients for reconstruction Fills wavelet fields: name, length and coefficients. Wavelet family Number of taps Computes orthonormal coefficients (from LoD coefficients only). Setups Haar wavelet. Setups Daubechies wavelet. Number of taps Setups Symlet wavelet. Number of taps Setups Coiflet wavelet. Number of taps Represents fractional delay line. Gets or sets interpolation mode. Gets the size of delay line (number of samples). Delay line. Current write position. Previously interpolated sample (used with InterpolationMode.Thiran). Constructs and reserves given for its samples. Delay line size (number of samples) Interpolation mode Constructs and reserves the size corresponding to seconds. Sampling rate Max delay (in seconds) Interpolation mode Writes (puts) to the delay line. Reads (gets) sample from the delay line corresponding to given time (in seconds). Resets delay line. Resizes delay line to ensure new . If does not exceed current size of the delay line then nothing happens. Resizes delay line to ensure new size corresponding to seconds. If the new size does not exceed current size of the delay line then nothing happens. Sampling rate Max delay (in seconds) Contains the most widely used contracts / guard clauses. Guards against negative number or zero. Argument (number) Argument name Guards against inequality of two arguments. The first argument The second argument Name of the first argument Name of the second argument Guards against the number being not in the given range. Argument (number) Lower boundary of the range Upper boundary of the range Argument name Guards against the case when the first and the second arguments are not valid boundaries of a range. The first argument The second argument Name of the first argument Name of the second argument Guards against the first argument exceeding the second argument. The first argument The second argument Name of the first argument Name of the second argument Guards against integer number being not power of 2 (e.g. 8, 16, 128, etc.) Argument (number) Argument name Guards against even integer number. Argument (number) Argument name Guards against not ordered and not unique array. Argument (array of values) Argument name Guards against incorrect parameters for (equiripple) filter design. Frequencies Desired magnitude response (gains) Weights Provides functions related to Linear Predictive Coding (LPC). Evaluates LP coefficients using Levinson-Durbin algorithm and returns prediction error. Auto-correlation vector LP coefficients Order of LPC Optional offset in auto-correlation vector Converts LPC coefficients to LPC cepstrum (LPCC). LPC vector Gain LPC cepstrum Converts LPC cepstrum to LPC coefficients and returns gain. LPC cepstrum LPC vector Estimates LPC order for a given according to the best practices. Sampling rate Converts LPC coefficients to Line Spectral Frequencies . The length of must be equal to length. Last element will be PI. LPC vector Line spectral frequencies Converts Line Spectral Frequencies to LPC coefficients. The length of must be equal to length. Last element must be PI. Line spectral frequencies LPC vector Provides helpful math functions. Returns Sinc of . Returns next power of 2 closest to the given number . Finds Greatest Common Divisor. Modulo function that works correctly with negative numbers (as np.mod). Computes Inverse Sinh of . Computes factorial !. Evaluates Binomial coefficient. Evaluate discrete difference of (array of the 1st order derivatives). Does linear interpolation (as numpy.interp). Does bilinear transform (in-place). Real parts of complex values Imaginary parts of complex values Unwraps phase. Phase array Jump size Wraps phase. Phase array Jump size Finds -th order statistics (n-th smallest value in array ). Modified Bessel function I0() of the 1st kind (using Taylor series, not very precise method). Number of iterations in Durand-Kerner algorithm for evaluating polynomial roots. Evaluates complex roots of polynomials using Durand-Kerner algorithm. Works for polynomials of order up to approx. 50. Polynomial coefficients Max number of iterations Checks if two arrays of complex numbers are essentially identical. Evaluates polynomial according to Horner scheme. Polynomial coefficients Argument Multiplies polynomials. Divides polynomials. Represents 2D matrix. Gets or sets number of rows. Gets or sets number of columns. Constructs with given number of and . Gets reference to underlying 2D array. Gets transposed matrix. Returns companion matrix. Input array Returns identity matrix of given . Returns sum of matrices and . Subtracts matrix from matrix . Gets row by its index. Row index Provides extension methods implementing fast operations with memory buffers. Creates array of single-precision values from enumerable of double-precision values. Creates array of double-precision values from enumerable of single-precision values. Creates fast copy of array. Makes fast copy of array (or its part) to existing array (or its part). Source array Destination array Number of elements to copy Offset in source array Offset in destination array Makes fast copy of array fragment starting at specified offset. Source array Number of elements to copy Offset in source array Offset in destination array Performs fast merging of array with array. Creates new array containing given array repeated times. Creates new zero-padded array of given from given array. Creates fast copy of array. Makes fast copy of array (or its part) to existing array (or its part). Source array Destination array Number of elements to copy Offset in source array Offset in destination array Makes fast copy of array fragment starting at specified offset. Source array Number of elements to copy Offset in source array Offset in destination array Performs fast merging of array with array. Creates new array containing given array repeated times. Creates new zero-padded array of given from given array. Provides methods for 1) converting between different scales: decibel MIDI pitch mel (HTK) mel (Slaney) bark1 (Traunmueller) bark2 (Wang) ERB 2) loudness weighting: A-weighting B-weighting C-weighting Converts magnitude value to dB level. Magnitude Reference magnitude Converts magnitude value to dB level (simplified version). Magnitude Converts power to dB level. Power Reference power Converts dB level to magnitude value. Decibel level Reference magnitude Converts dB level to magnitude value (simplified version). Decibel level Converts dB level to power. Decibel level Reference power Converts MIDI pitch to frequency (in Hz). Pitch Converts frequency to MIDI pitch. Frequency (in Hz) Array of musical notes. Converts musical note (in format ("G", 3), ("E", 5), etc.) to frequency in Hz. Note (A-G#) Octave (0-8) Converts frequency in Hz to note (in tuple format ("G", 3), ("E", 5), etc.). Frequency in Hz Converts herz frequency to corresponding mel frequency. Converts mel frequency to corresponding herz frequency. Converts herz frequency to mel frequency (suggested by M.Slaney). Converts mel frequency to herz frequency (suggested by M.Slaney). Converts herz frequency to corresponding bark frequency (according to Traunmüller (1990)). Converts bark frequency to corresponding herz frequency (according to Traunmüller (1990)). Converts herz frequency to corresponding bark frequency (according to Wang (1992)); used in M.Slaney's auditory toolbox. Converts bark frequency to corresponding herz frequency (according to Wang (1992)); used in M.Slaney's auditory toolbox. Converts herz frequency to corresponding ERB frequency. Converts ERB frequency to corresponding herz frequency. Converts Hz frequency to octave (used for constructing librosa-like Chroma filterbanks). Returns perceptual loudness weight (in dB). Frequency Weighting type (A, B, C) Generates window coefficients of various types. Generates window coefficients of given and . Window type Window length Additional optional parameters Generates rectangular window of given . Generates triangular window of given . Generates Hamming window of given . Generates Blackman window of given . Generates Hann window of given . Generates Gaussian window of given . Generates Kaiser window of given . Generates Kaiser-Bessel Derived window of given . Generates Bartlett-Hann window of given . Generates Lanczos window of given . Generates Sin-beta window of given . Generates Flat-top window of given . Generates coefficients for cepstrum liftering. Length of the window Denominator in liftering formula Provides extension methods for applying windows to signals and arrays of samples. Applies window to array of . Samples Window coefficients Applies window to array of . Samples Window coefficients Applies window to . Signal Window coefficients Applies window with optional to array of . Samples Window type Window parameters Applies window with optional to array of . Samples Window type Window parameters Applies window with optional to . Signal Window type Window parameters Defines the most commonly used window functions. Rectangular window. Triangular window. Hamming window. Blackman window. Hann window. Gaussian window. Kaiser window. Kaiser-Bessel Derived window. Bartlett-Hann window. Lanczos window. Power-of-sine window. Flat-top window. Window for cepstral liftering.