krotos::AudioEmbedding Class Reference

#include <AudioEmbedding.h>

Public Member Functions
	AudioEmbedding ()
void	initialise ()
std::vector< float >	forward (const AudioSampleBuffer &buffer)
std::vector< float >	hz_to_mel (std::vector< float > freqs, bool htk=false)
std::vector< float >	mel_to_hz (std::vector< float > mels, bool htk=false)
std::vector< float >	mel_frequencies (float fmin, float fmax, int n_mels, bool htk=false)
std::vector< std::vector< float > >	mel (int sr, int n_fft, int n_mels, float fmin, float fmax, bool htk=false)

Private Member Functions
void	L2Normalise (std::vector< float > &x, float eps=1e-5f)
std::vector< float >	applyMelFilterbank (const std::vector< float > &x)
std::vector< float >	applyAutoEncoder (const std::vector< float > &x)
void	writeMatrixToFile (const std::vector< std::vector< float > > &matrix, const std::string &filename)

Private Attributes
bool	m_initialised
int	m_fftSize
int	m_hopSize
juce::dsp::FFT	m_fft
juce::dsp::WindowingFunction< float >	m_window
std::vector< std::vector< float > >	m_mel
std::vector< std::vector< float > >	m_weight

Constructor & Destructor Documentation

AudioEmbedding()

krotos::AudioEmbedding::AudioEmbedding

(

)

Constructor

Member Function Documentation

applyAutoEncoder()

std::vector< float > krotos::AudioEmbedding::applyAutoEncoder ( const std::vector< float > & x )

private

Applies an autoencoder to the input.

Parameters

x	input to autoencoder

applyMelFilterbank()

std::vector< float > krotos::AudioEmbedding::applyMelFilterbank ( const std::vector< float > & x )

private

Applies a filterbank to the input.

Parameters

x	input to apply filterbank

forward()

std::vector< float > krotos::AudioEmbedding::forward

(

const AudioSampleBuffer &

buffer

)

Computes an audio embedding.

Parameters

buffer

buffer of audio samples to analyse

Returns

audio embedding vector

hz_to_mel()

std::vector< float > krotos::AudioEmbedding::hz_to_mel	(	std::vector< float >	freqs,
		bool	htk = false )

Convert Hz to Mels.

Parameters

freqs	vector of frequencies
htk	use HTK formula instead of Slaney

Returns

input frequencies in Mels

initialise()

void krotos::AudioEmbedding::initialise

(

)

Initialise with default parameters

L2Normalise()

void krotos::AudioEmbedding::L2Normalise ( std::vector< float > & x, float eps = 1e-5f )

private

Normalises a vector using the L2 norm.

Parameters

x	vector to normalise
eps	epsilon to avoid division by zero

mel()

std::vector< std::vector< float > > krotos::AudioEmbedding::mel	(	int	sr,
		int	n_fft,
		int	n_mels,
		float	fmin,
		float	fmax,
		bool	htk = false )

Create a Mel filter-bank.

Parameters

sr	sampling rate of the incoming signal
n_fft	number of FFT components
n_mels	number of mel bins
fmin	minimum frequency (Hz).
fmax	maximum frequency (Hz).
htk	use HTK formula instead of Slaney

Returns

Mel transform matrix

mel_frequencies()

std::vector< float > krotos::AudioEmbedding::mel_frequencies	(	float	fmin,
		float	fmax,
		int	n_mels,
		bool	htk = false )

Compute an array of acoustic frequencies tuned to the mel scale.

Parameters

n_mels	number of mel bins
fmin	minimum frequency (Hz).
fmax	maximum frequency (Hz).
htk	if True, use HTK formula to convert Hz to mel. Otherwise (False), use Slaney’s Auditory Toolbox

Returns

vector of n_mels frequencies in Hz which are uniformly spaced on the Mel axis

mel_to_hz()

std::vector< float > krotos::AudioEmbedding::mel_to_hz	(	std::vector< float >	mels,
		bool	htk = false )

Convert mel bin numbers to frequencies.

Parameters

mels	mel bins to convert
htk	use HTK formula instead of Slaney

Returns

input mels in Hz

writeMatrixToFile()

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

private

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

Member Data Documentation

m_fft

juce::dsp::FFT krotos::AudioEmbedding::m_fft

private

m_fftSize

int krotos::AudioEmbedding::m_fftSize

private

m_hopSize

int krotos::AudioEmbedding::m_hopSize

private

m_initialised

bool krotos::AudioEmbedding::m_initialised

private

m_mel

std::vector<std::vector<float> > krotos::AudioEmbedding::m_mel

private

m_weight

std::vector<std::vector<float> > krotos::AudioEmbedding::m_weight

private

linear autoencoder weights exported from PyTorch

m_window

juce::dsp::WindowingFunction<float> krotos::AudioEmbedding::m_window

private

---

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

• krotos_ml/ucs_search/AudioEmbedding.h
• krotos_ml/ucs_search/AudioEmbedding.cpp