com.tagtraum.jipes.audio

Class MelSpectrum

java.lang.Object

com.tagtraum.jipes.audio.AbstractAudioSpectrum

com.tagtraum.jipes.audio.MelSpectrum

All Implemented Interfaces:

AudioBuffer, AudioSpectrum, Cloneable

public class MelSpectrum
extends AbstractAudioSpectrum
implements Cloneable

Mel spectrum typically constructed from a LinearFrequencySpectrum created by FFT. Each mel channel is computed by applying a triangular filter to the magnitudes or powers of the input spectrum and adding up the result. The filters for consecutive channels overlap by 0.5 of their length.
Computing the mel spectrum from the magnitudes happens similar to the specification in European Telecommunications Standards Institute (2003), Speech Processing, Transmission and Quality Aspects (STQ); Distributed speech recognition; Front-end feature extraction algorithm; Compression algorithms. Technical standard ES 201 108, v1.1.3.
The European Telecommunications Standards Institute requires the filtering of the magnitudes. However, this implementation lets you choose between magnitude and power filtering.

Author:

Hendrik Schreiber

Field Summary

Fields inherited from class com.tagtraum.jipes.audio.AbstractAudioSpectrum

audioFormat, frameNumber, imaginaryData, magnitudes, powers, realData

Constructor Summary

Constructors

Constructor and Description
MelSpectrum(int frameNumber, float[] real, float[] imaginary, AudioFormat audioFormat, float[][] filterBank, float[] channelBoundariesInHz)
MelSpectrum(int frameNumber, LinearFrequencySpectrum audioSpectrum, float[][] filterBank, float[] channelBoundariesInHz, boolean filterPowers) Creates a mel spectrum from a linear spectrum.
MelSpectrum(int frameNumber, LinearFrequencySpectrum audioSpectrum, float lowerFrequency, float upperFrequency, int channels, boolean filterPowers) Creates a mel spectrum from a linear spectrum.
MelSpectrum(MelSpectrum melSpectrum)

Method Summary

Modifier and Type	Method and Description
static float[]	channelBoundaries(float lowerFrequency, float upperFrequency, int boundaryCount) Creates an array with boundaryCount frequencies in Hz, dividing the given frequency range into boundaryCount - 1 bins with equal mel-length.
Object	clone() Creates an exact copy of this buffer.
static float[][]	createFilterBank(float[] frequencies, float[] channelBoundariesInHz) Creates an array of filter coefficients.
MelSpectrum	derive(float[] real, float[] imaginary) Creates a copy of this spectrum, but replaces its values with the given real and imaginary data.
boolean	equals(Object o)
int	getBin(float frequency) Computes a bin for the given frequency.
float[]	getFrequencies() Array with frequency values in Hz corresponding to the bin numbers.
int	getNumberOfBands() Number of channels/bands.
int	hashCode()
String	toString()

Methods inherited from class com.tagtraum.jipes.audio.AbstractAudioSpectrum

computeMagnitudes, computePowers, getAudioFormat, getData, getFrameNumber, getFrequency, getImaginaryData, getMagnitudes, getNumberOfSamples, getPowers, getRealData, getTimestamp, getTimestamp, reuse

Methods inherited from class java.lang.Object

finalize, getClass, notify, notifyAll, wait, wait, wait

Constructor Detail

MelSpectrum

public MelSpectrum(int frameNumber,
                   LinearFrequencySpectrum audioSpectrum,
                   float lowerFrequency,
                   float upperFrequency,
                   int channels,
                   boolean filterPowers)

Creates a mel spectrum from a linear spectrum.
Note that the upper and lower boundaries are not the central frequencies of the mel channels, but rather their edges.

Parameters:

frameNumber - frame number

audioSpectrum - audio spectrum to copy data from

lowerFrequency - lower frequency boundary for the lowest triangular filter in Hz

upperFrequency - upper frequency boundary for the highest triangular filter in Hz

channels - number of mel channels

filterPowers - if true, the power spectrum is filtered, if false, the magnitude spectrum is filtered

MelSpectrum

public MelSpectrum(int frameNumber,
                   LinearFrequencySpectrum audioSpectrum,
                   float[][] filterBank,
                   float[] channelBoundariesInHz,
                   boolean filterPowers)

Creates a mel spectrum from a linear spectrum.

Parameters:

frameNumber - frame number

audioSpectrum - audio spectrum to copy data from

filterBank - filterbank with coefficients for each channel

channelBoundariesInHz - channel boundaries in Hz

filterPowers - if true, the power spectrum is filtered, if false, the magnitude spectrum is filtered

See Also:

createFilterBank(float[], float[]), channelBoundaries(float, float, int)

MelSpectrum

public MelSpectrum(MelSpectrum melSpectrum)

MelSpectrum

public MelSpectrum(int frameNumber,
                   float[] real,
                   float[] imaginary,
                   AudioFormat audioFormat,
                   float[][] filterBank,
                   float[] channelBoundariesInHz)

Method Detail

channelBoundaries

public static float[] channelBoundaries(float lowerFrequency,
                                        float upperFrequency,
                                        int boundaryCount)

Creates an array with boundaryCount frequencies in Hz, dividing the given frequency range into boundaryCount - 1 bins with equal mel-length.

Parameters:

lowerFrequency - lowerFrequency in Hz

upperFrequency - upperFrequency in Hz

boundaryCount - number of boundaries

Returns:

array with boundary frequencies

createFilterBank

public static float[][] createFilterBank(float[] frequencies,
                                         float[] channelBoundariesInHz)

Creates an array of filter coefficients.

Parameters:

frequencies - frequencies for a value input array

channelBoundariesInHz - channel boundaries

Returns:

filter bank

derive

public MelSpectrum derive(float[] real,
                          float[] imaginary)

Description copied from interface: AudioSpectrum

Creates a copy of this spectrum, but replaces its values with the given real and imaginary data. The original object remains unchanged.

Specified by:

derive in interface AudioBuffer

Specified by:

derive in interface AudioSpectrum

Parameters:

real - real data

imaginary - imaginary data

Returns:

derived copy

getNumberOfBands

public int getNumberOfBands()

Number of channels/bands.

Returns:

number of channels/bands

getBin

public int getBin(float frequency)

Computes a bin for the given frequency. Returns -1, if the bin is not found.

Specified by:

getBin in interface AudioSpectrum

Parameters:

frequency - frequency

Returns:

bin

getFrequencies

public float[] getFrequencies()

Description copied from interface: AudioSpectrum

Array with frequency values in Hz corresponding to the bin numbers.

This array is meant for reading only. Do not manipulate it!

Specified by:

getFrequencies in interface AudioSpectrum

Returns:

array with frequencies - length may be numberOfSamples/2 due to symmetry

See Also:

AudioSpectrum.getFrequency(int)

equals

public boolean equals(Object o)

Overrides:

equals in class Object

hashCode

public int hashCode()

Overrides:

hashCode in class Object

toString

public String toString()

Overrides:

toString in class Object

clone

public Object clone()
             throws CloneNotSupportedException

Description copied from interface: AudioBuffer

Creates an exact copy of this buffer. All AudioBuffers must be cloneable.

Specified by:

clone in interface AudioBuffer

Overrides:

clone in class AbstractAudioSpectrum

Returns:

clone

Throws:

CloneNotSupportedException