FFT-based cochlear power spectrogram. More...

#include <ERB_FFTSpectrogram.h>

Public Member Functions
	ERB_FFTSpectrogram ()

void	setSampleRate (float fs)

std::vector< std::vector< float > >	filterSpectrum (std::vector< float > &inputSignal)

void	drawSpectrogram (juce::Image &image, const std::vector< std::vector< float > > &erbSTFTPowMatrix)

std::vector< float >	getERBSpace ()

Static Public Member Functions
static std::vector< float >	erbs2Hz (const std::vector< float > &erbsVector)

static Eigen::VectorXf	erbs2Hz (const Eigen::VectorXf &erbsVector)

static std::vector< float >	hz2Erbs (const std::vector< float > &hzVector)

static Eigen::VectorXf	hz2Erbs (const Eigen::VectorXf &hzVector)

static float	hz2Erbs (const float hz)

Private Member Functions
std::vector< float >	applyEarFilter (std::vector< float > &inputSignal)

std::vector< float >	gtWindow (int numberOfPoints, float bParam, int order)

std::vector< float >	cferb (std::vector< float > cf)

float	cferb (const float cf)

double	cferb (const double cf)

std::vector< float >	erbSpace (float low, float hi, int N)

float	centroid (const std::vector< float > &x)

float	getNextPowerOfTwo (float value)

void	buildFilterBank (int startSamplesSize)

std::vector< std::vector< float > >	calculatePowerSpectrum (const std::vector< std::vector< float > > &inFrames)

void	writeMatrixToFile (const std::vector< std::vector< float > > &matrix, const std::string &filename)

Private Attributes
float	m_sampleRate {-1.0f}

std::vector< std::vector< double > >	m_wfunct

std::vector< float >	m_cfArray

EarFilter	m_earFilter

float	m_hopSizeSec = 0.0058f

int	m_hopSizeSamples

const float	m_exponent = 0.25f

float	m_lo = 30.0f

float	m_hi = 16000.0f

int	m_numChans = 77

float	m_bw0 = 24.7f

float	m_b0 = 25.15274949f

float	m_ERD = 0.0196797f

int	m_wSize

std::vector< float >	m_gtWin

const float	m_scalingConstant = 0.1f

int	m_fftSize

Detailed Description

FFT-based cochlear power spectrogram.

Power spectrogram with same frequency resolution and scale as human ear. ERB_FFTSpectrogram splits the signal into overlapping frames and windows each with a function shaped like the time-reversed envelope of a gammatone impulse response with parameters appropriate for a low-frequency cochlear filter, with an equivalent rectangular duration (ERD) of about 20ms. The FFT size is set to the power of two immediately larger than twice that value (sr*0.040), and the windowed slices are Fourier transformed to obtain a power spectrum. The frequency resolution is that of a "cf=0 Hz" channel, ie narrower than even the lowest cf channels, and the frequency axis is linear. To get a resolution similar to the cochlea, and channels evenly spaced on an equal-resolution scale (Cambridge ERBs), the power spectrum is remapped. Each channel of the new spectrum is the weighted sum of power spectrum coefficients, obtained by forming the vector product with a weighting function so that the channel has its proper spectral width.

Constructor & Destructor Documentation

◆ ERB_FFTSpectrogram()

krotos::ERB_FFTSpectrogram::ERB_FFTSpectrogram ( )

Constructor

Copyright Krotos LTD 2023

Member Function Documentation

◆ applyEarFilter()

std::vector< float > krotos::ERB_FFTSpectrogram::applyEarFilter ( std::vector< float > & inputSignal )

private

Applies the outer/middle ear filter using the EarFilter object

Parameters

inputSignal vector representing the input Signal

Returns: effective signal at entrance of cochlear filter bank

◆ buildFilterBank()

void krotos::ERB_FFTSpectrogram::buildFilterBank ( int startSamplesSize )

private

Builds the ERB scale Filterbank - saves result in m_wfunct. Casted to double to rectify float underflow!!

Parameters

startSampleSize last hop index where the window will fit within the end of the signal

◆ calculatePowerSpectrum()

std::vector< std::vector< float > > krotos::ERB_FFTSpectrogram::calculatePowerSpectrum ( const std::vector< std::vector< float > > & inFrames )

private

Calculates the power spectrum per column

Parameters

inFrames matrix represting the overlapping window frames of signal

Returns: powerSpectrum of signal TODO:: In the future it can be improved a lot for performance and also call ShortTimeFourierTransform to perform

◆ centroid()

float krotos::ERB_FFTSpectrogram::centroid ( const std::vector< float > & x )

private

Centroid of matrix x

Parameters

x matrix

Returns: centroid as weighted average of row index

◆ cferb() [1/3]

double krotos::ERB_FFTSpectrogram::cferb ( const double cf )

private

◆ cferb() [2/3]

float krotos::ERB_FFTSpectrogram::cferb ( const float cf )

private

◆ cferb() [3/3]

std::vector< float > krotos::ERB_FFTSpectrogram::cferb ( std::vector< float > cf )

private

◆ drawSpectrogram()

void krotos::ERB_FFTSpectrogram::drawSpectrogram	(	juce::Image &	image,
		const std::vector< std::vector< float > > &	erbSTFTPowMatrix )

Draws a simple spectrogram for POC purposes. Assumes we have already produced the STFT matrix using the filterSpectrum() above.

Parameters

image	the Image component whose pixels we want to colour according to the mag of the bins. TODO: write a better scaling algorithm and average the signal magnitudes.
stftMatrix	the STFT matrix in mxn dimensions where m: mth frame, n: nth bin. NOTE: you should call repaint after this so that the new Image appears.

◆ erbs2Hz() [1/2]

Eigen::VectorXf krotos::ERB_FFTSpectrogram::erbs2Hz ( const Eigen::VectorXf & erbsVector )

static

◆ erbs2Hz() [2/2]

std::vector< float > krotos::ERB_FFTSpectrogram::erbs2Hz ( const std::vector< float > & erbsVector )

static

Create ERB - scale uniformly - spaced frequencies(in Hertz)

◆ erbSpace()

std::vector< float > krotos::ERB_FFTSpectrogram::erbSpace	(	float	low,
		float	hi,
		int	N )

private

Computes an array of N frequencies uniformly spaced between lo and hi in the ERB scale. For a definition of ERB, see Moore, B. C. J., and Glasberg, B. R. (1983). "Suggested formulae for calculating auditory-filter bandwidths and excitation patterns," J. Acoust. Soc. Am. 74, 750-753.

Parameters

low	Hz - lower limit of vector
hi	Hz - upper limit of vector
N	number of values to produce

Returns: vector of values

◆ filterSpectrum()

std::vector< std::vector< float > > krotos::ERB_FFTSpectrogram::filterSpectrum ( std::vector< float > & inputSignal )

Performs the fft power filtering operation

Parameters

inputSignal vector representing the input Signal

Returns: vector of vectors where each column is a different frame and each row a frequency bin.

◆ getERBSpace()

std::vector< float > krotos::ERB_FFTSpectrogram::getERBSpace ( )

inline

◆ getNextPowerOfTwo()

float krotos::ERB_FFTSpectrogram::getNextPowerOfTwo ( float value )

private

Next power of two of float

Parameters

value number whose nextpow2 I'm looking for

Returns: nextpow2

◆ gtWindow()

std::vector< float > krotos::ERB_FFTSpectrogram::gtWindow	(	int	numberOfPoints,
		float	bParam,
		int	order )

private

Caclulates a window shaped like a time-reversed gammatone envelope

Parameters

numberOfPoints	number of points
bParam	b-parameter
order	order of function

◆ hz2Erbs() [1/3]

Eigen::VectorXf krotos::ERB_FFTSpectrogram::hz2Erbs ( const Eigen::VectorXf & hzVector )

static

◆ hz2Erbs() [2/3]

float krotos::ERB_FFTSpectrogram::hz2Erbs ( const float hz )

static

◆ hz2Erbs() [3/3]

std::vector< float > krotos::ERB_FFTSpectrogram::hz2Erbs ( const std::vector< float > & hzVector )

static

Create frequency vector in Hz from the ERB equivalent vector

◆ setSampleRate()

void krotos::ERB_FFTSpectrogram::setSampleRate ( float fs )

Sets the sampling Rate

Parameters

fs	sampling Rate [Hz]

◆ writeMatrixToFile()

void krotos::ERB_FFTSpectrogram::writeMatrixToFile	(	const std::vector< std::vector< float > > &	matrix,
		const std::string &	filename )

inlineprivate

This function exists as a debugging mechanism to compare the spectrogram created here with it's equivalent in MATLAB.

Member Data Documentation

◆ m_b0

float krotos::ERB_FFTSpectrogram::m_b0 = 25.15274949f

private

◆ m_bw0

float krotos::ERB_FFTSpectrogram::m_bw0 = 24.7f

private

◆ m_cfArray

std::vector<float> krotos::ERB_FFTSpectrogram::m_cfArray

private

◆ m_earFilter

EarFilter krotos::ERB_FFTSpectrogram::m_earFilter

private

◆ m_ERD

float krotos::ERB_FFTSpectrogram::m_ERD = 0.0196797f

private

◆ m_exponent

const float krotos::ERB_FFTSpectrogram::m_exponent = 0.25f

private

◆ m_fftSize

int krotos::ERB_FFTSpectrogram::m_fftSize

private

◆ m_gtWin

std::vector<float> krotos::ERB_FFTSpectrogram::m_gtWin

private

◆ m_hi

float krotos::ERB_FFTSpectrogram::m_hi = 16000.0f

private

◆ m_hopSizeSamples

int krotos::ERB_FFTSpectrogram::m_hopSizeSamples

private

◆ m_hopSizeSec

float krotos::ERB_FFTSpectrogram::m_hopSizeSec = 0.0058f

private

◆ m_lo

float krotos::ERB_FFTSpectrogram::m_lo = 30.0f

private

◆ m_numChans

int krotos::ERB_FFTSpectrogram::m_numChans = 77

private

◆ m_sampleRate

float krotos::ERB_FFTSpectrogram::m_sampleRate {-1.0f}

private

◆ m_scalingConstant

const float krotos::ERB_FFTSpectrogram::m_scalingConstant = 0.1f

private

◆ m_wfunct

std::vector<std::vector<double> > krotos::ERB_FFTSpectrogram::m_wfunct

private

◆ m_wSize

int krotos::ERB_FFTSpectrogram::m_wSize

private

The documentation for this class was generated from the following files:

krotos_dsp/utilities/analysis_modules/frequency_domain_analysis/ERB_FFTSpectrogram.h
krotos_dsp/utilities/analysis_modules/frequency_domain_analysis/ERB_FFTSpectrogram.cpp

Public Member Functions

Static Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

◆ ERB_FFTSpectrogram()

Member Function Documentation

◆ applyEarFilter()

◆ buildFilterBank()

◆ calculatePowerSpectrum()

◆ centroid()

◆ cferb() [1/3]

◆ cferb() [2/3]

◆ cferb() [3/3]

◆ drawSpectrogram()

◆ erbs2Hz() [1/2]

◆ erbs2Hz() [2/2]

◆ erbSpace()

◆ filterSpectrum()

◆ getERBSpace()

◆ getNextPowerOfTwo()

◆ gtWindow()

◆ hz2Erbs() [1/3]

◆ hz2Erbs() [2/3]

◆ hz2Erbs() [3/3]

◆ setSampleRate()

◆ writeMatrixToFile()

Member Data Documentation

◆ m_b0

◆ m_bw0

◆ m_cfArray

◆ m_earFilter

◆ m_ERD

◆ m_exponent

◆ m_fftSize

◆ m_gtWin

◆ m_hi

◆ m_hopSizeSamples

◆ m_hopSizeSec

◆ m_lo

◆ m_numChans

◆ m_sampleRate

◆ m_scalingConstant

◆ m_wfunct

◆ m_wSize