org.openimaj.audio.filters

Class MelFilterBank

java.lang.Object

org.openimaj.audio.filters.MelFilterBank

public class MelFilterBank
extends Object

Filter bank of Mel filters for applying to a frequency domain source. It is standard that the edges of each filter in the filter bank correspond to the centre of the neighbouring filter - so they overlap by half (in the Mel frequencies).

Version:

$Author$, $Revision$, $Date$

Author:

David Dupplaw (dpd@ecs.soton.ac.uk)

Created:

25 Jul 2012

Constructor Summary

Constructors

Constructor and Description
MelFilterBank() Construct a default MelFilterBank.
MelFilterBank(int nFilters, double lowFreq, double highFreq) Default constructor to create a filter bank with the given number of filters between the two given frequencies.

Method Summary

Modifier and Type	Method and Description
Jama.Matrix	asMatrix(int specSize, double minFreq, double maxFreq) Returns the filters weights as a Matrix, where the rows are the filters and the columns are the frequencies.
void	createFilterBank() Instantiate the filter bank, if it's not already instantiated.
List<TriangularFilter>	getFilters() Returns a list of filters in this filter bank
float[]	getResponseCurve(int nSpectrumBins, double maxFreq) Returns a set of values that represent the response of this filter bank when the linear frequency is split in the given number of bins.
float[][]	process(float[][] spectrum, AudioFormat format) Process the input power spectrum with this filter bank.
void	setFilterAmplitude(double fa) Set the filter amplitude for all the generated filters.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

MelFilterBank

public MelFilterBank()

Construct a default MelFilterBank. The defaults are the lowest frequency covered is 300Hz, the highest 5000Hz covered by 40 Mel filters.

MelFilterBank

public MelFilterBank(int nFilters,
                     double lowFreq,
                     double highFreq)

Default constructor to create a filter bank with the given number of filters between the two given frequencies.

Parameters:

nFilters - The number of filters

lowFreq - The lowest frequency covered by the bank

highFreq - The highest frequency covered by the bank

Method Detail

createFilterBank

public void createFilterBank()

Instantiate the filter bank, if it's not already instantiated.

getFilters

public List<TriangularFilter> getFilters()

Returns a list of filters in this filter bank

Returns:

The filters

process

public float[][] process(float[][] spectrum,
                         AudioFormat format)

Process the input power spectrum with this filter bank. The output is a set of Mel Frequency Coefficients for each channel of the audio. The power spectrum is expected to be just the power magnitudes for the real parts of a fourier frequency spectrum (for each channel of the input audio). The output is a 2D array where the first dimension is the number of audio channels and the second dimension is each of the powers of the mel filters.

Parameters:

spectrum - The power spectrum

format - The format of the original audio used to produce the spectrum

Returns:

The Mel frequency coefficients

getResponseCurve

public float[] getResponseCurve(int nSpectrumBins,
                                double maxFreq)

Returns a set of values that represent the response of this filter bank when the linear frequency is split in the given number of bins. The result will have nSpectrumBins length.

Parameters:

nSpectrumBins - The number of bins in a spectrum.

maxFreq - The maximum frequency (sample rate)

Returns:

The response curve.

setFilterAmplitude

public void setFilterAmplitude(double fa)

Set the filter amplitude for all the generated filters.

Parameters:

fa - The new filter amplitude

asMatrix

public Jama.Matrix asMatrix(int specSize,
                            double minFreq,
                            double maxFreq)

Returns the filters weights as a Matrix, where the rows are the filters and the columns are the frequencies. The values in the cells are the weights to apply to the given frequency for the given filter. The frequency range of the spectrum is required to work out which frequency each bin represents.

Parameters:

specSize - The size of the spectrum (number of cols in the matrix)

minFreq - The minimum frequency represented in the spectrum

maxFreq - The maximum frequency represented in the spectrum

Returns:

a Matrix representation of the filter bank or NULL if the filter bank has not yet been initialised