MFCC Class Reference

Algorithm to calculate the Mel-frequency Cepstrum Coefficients of vectors of Real values. More...

#include <MFCC.h>

Collaboration diagram for MFCC:

[legend]

List of all members.

Public Member Functions
	MFCC (Real lowFrequency=300.0, Real highFrequency=16000.0, int bandCount=40.0, Real samplerate=44100.0, int fftSize=1024, int coefficientCount=13, Real minSpectrum=1e-10, Real power=1.0)
	~MFCC ()
void	process (const MatrixXR &spectrums, MatrixXR *mfccCoefficients)
int	coefficientCount () const
void	setCoefficientCount (int count, bool callSetup=true)
int	bandCount () const
void	setBandCount (int count, bool callSetup=true)
Real	lowFrequency () const
void	setLowFrequency (Real frequency, bool callSetup=true)
Real	highFrequency () const
void	setHighFrequency (Real frequency, bool callSetup=true)
Real	samplerate () const
void	setSamplerate (Real frequency, bool callSetup=true)
int	fftSize () const
void	setFftSize (int size, bool callSetup=true)
Real	power () const
void	setPower (Real factor, bool callSetup=true)

---

Detailed Description

Algorithm to calculate the Mel-frequency Cepstrum Coefficients of vectors of Real values.

This class represents an object to perform a Mel-frequency Cepstrum Coefficients (MFCC) on vectors of Real values. Which is a useful technique creating a sparse representation of a spectrum magnitude. The algorithm estimates a set of M coefficients which should be similar for perceptually similar sounds.

The MFCCs are calculated by taking the Discrete Cosine Transform (DCT) of the values of the Mel log powers. The Mel log powers are calculated by applying the logarithm to the vaules of the Mel Bands (MelBands). The Mel Bands are triangular overlapping windows applied on the power spectrum mapped onto the Mel-frequency scale.

The samplerate and FFT size of the input spectrum are specified using setSamplerate() and setFftSize().

The frequency limits of the Mel scale mapping are specified using setLowFrequency() and setHighFrequency().

The number of Mel bands is specified using setBandCount().

The number of resulting DCT coefficients is specified by setCoefficientCount().

Author:

Ricard Marxer

See also:

MelBands, Bands, LPC

---

Constructor & Destructor Documentation

MFCC	(	Real	lowFrequency = 300.0,
		Real	highFrequency = 16000.0,
		int	bandCount = 40.0,
		Real	samplerate = 44100.0,
		int	fftSize = 1024,
		int	coefficientCount = 13,
		Real	minSpectrum = 1e-10,
		Real	power = 1.0
	)

Constructs an MFCC object with the specified lowFrequency, highFrequency, bandCount, samplerate, fftSize, coefficientCount, minSpectrum and power settings.

Parameters:

	lowFrequency	frequency of the lowest Mel band, must be greater than zero 0 and lower than half the samplerate.
	highFrequency	frequency of the highest Mel band, must be greater than zero 0 and lower than half the samplerate.
	bandCount	number of Mel bands.
	samplerate	samplerate frequency of the input signal.
	fftSize	size of the FFT.
	coefficientCount	number of DCT coefficients to be estimated.
	minSpectrum	value to which the spectrum is clipped before performing the logarithm.
	power	value to which to power the band values before performing the DCT.

References setBandCount(), setCoefficientCount(), setFftSize(), setHighFrequency(), setLowFrequency(), setPower(), and setSamplerate().

~MFCC

(

)

Destroys the algorithm and frees its resources.

---

Member Function Documentation

void process	(	const MatrixXR &	spectrums,
		MatrixXR *	mfccCoefficients
	)

Performs an MFCC on each of the rows of frames. Puts the resulting MFCC coefficients in the rows of mfccCoefficients.

Parameters:

	spectrums	matrix of Real values representing one spectrum magnitude per row. The number of columns of spectrum must be equal to the fftSize / 2 + 1 where fftSize is specified using setFftSize().
	mfccCoefficients	pointer to a matrix of Real values for the MFCC coefficients. The matrix should have the same number of rows as spectrums and coefficientCount columns.

Note that if the output matrices are not of the required sizes they will be resized, reallocating a new memory space if necessary.

References DCT::process(), and MelBands::process().

int coefficientCount

(

)

const

Returns the number of coefficients to be calculated. The default is 13.

See also:

setCoefficientCount()

void setCoefficientCount	(	int	count,
		bool	callSetup = true
	)

Specifies the count of coefficients to be calculated. The given count must be in the range between 0 and (input size - 1).

See also:

coefficientCount()

Referenced by MFCC().

int bandCount

(

)

const

Returns the number of bands to be performed. The default is 40.

See also:

setBandCount()

void setBandCount	(	int	count,
		bool	callSetup = true
	)

Specifies the count of bands to be performed.

See also:

bandCount()

Referenced by MFCC().

Real lowFrequency

(

)

const

Return the low frequency of the MFCC. The default is 300.0.

See also:

lowFrequency, highFrequency, setLowFrequency, setHighFrequency

void setLowFrequency	(	Real	frequency,
		bool	callSetup = true
	)

Specifies the low frequency of the MFCC. The given frequency must be in the range of 0 to the samplerate / 2.

See also:

lowFrequency, highFrequency, setHighFrequency

Referenced by MFCC().

Real highFrequency

(

)

const

Return the high frequency of the MFCC. The default is 16000.0.

See also:

lowFrequency, setLowFrequency, setHighFrequency

void setHighFrequency	(	Real	frequency,
		bool	callSetup = true
	)

Specifies the high frequency of the MFCC. The given frequency must be in the range of 0 to the samplerate / 2.

See also:

lowFrequency, highFrequency, setLowFrequency

Referenced by MFCC().

Real samplerate

(

)

const

Return the samplerate frequency of the input signal. The default is 44100.0.

See also:

setSamplerate

void setSamplerate	(	Real	frequency,
		bool	callSetup = true
	)

Specifies the samplerate frequency of the input signal.

See also:

samplerate

Referenced by MFCC().

int fftSize

(

)

const

Returns the size of the FFT that has been performed for the input. The default is 1024.

See also:

setFftSize()

void setFftSize	(	int	size,
		bool	callSetup = true
	)

Specifies the size of the FFT that has been performed for the input. The given size must be higher than 0.

See also:

fftSize()

Referenced by MFCC().

Real power

(

)

const

Return the factor to which the bands are powered. The default is 1.0.

See also:

setPower

void setPower	(	Real	factor,
		bool	callSetup = true
	)

Specifies the factor to which the bands are powered.

See also:

power

Referenced by MFCC().

---

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

• MFCC.h
• MFCC.cpp