This method is octave frequency scale equivalent of MFCCStuff.DecibelSpectra(dspOutput.AmplitudeSpectrogram, dspOutput.WindowPower, sampleRate, epsilon) The MFCCStuff method is the proper way to convert amplitude spectrogram to decibels. It converts an amplitude spectrogram to a power spectrogram having specified frequency scale. It transforms the amplitude spectrogram in the following steps: (1) It removes the DC row or bin 0 iff there is odd number of spectrogram bins. ASSUMPTION: Bin count should be power of 2 from FFT. (2) Then reduce the linear scale to an octave scale depending on the sr and required number of bins or filters. (3) It converts spectral amplitudes to power, normalising for window power and sample rate. The window contributes power to the signal which must subsequently be removed from the spectral power. Calculate power per sample. See notes in the MFCCStuff.DecibelSpectra for further exp[lanaitons. These normalisations were adapted from MatLab MFCC code. (4) It converts the power spectrogram to decibels.

Declaration

public static double[, ] ConvertAmplitudeSpectrogramToFreqScaledDecibels(double[, ] amplitudeM, double windowPower, int sampleRate, double epsilon, FrequencyScale freqScale)

Parameters

Type	Name	Description
Double[,]	amplitudeM	The amplitude spectrogram.
Double	windowPower	FFT window power comes from DSP_Frames.WindowPower.
Int32	sampleRate	of the original signal.
Double	epsilon	dependson bit-rate of the original signal.
FrequencyScale	freqScale	In this case an octave frquency scale.

Returns

Type	Description
Double[,]	The decibel spectrogram.

| Improve this Doc View Source

ConvertAmplitudeToPowerSpectrogram(Double[,], Double, Int32)

Converts Amplitude Spectrogram to Power Spectrogram. Square the amplitude values to get power. Power values must be adjusted for the power in the FFT window and also for the sample rate. Must divide by the window power to remove its contribution to amplitude values. Must divide by sample rate to get average power per signal sample. NOTE: Multiply by 2 to accomodate two spectral components, ie positive and neg freq. BUT the last nyquist bin does not require a factor of two. However this method is called only by octave reduced frequency scales where the nyquist bin is just one of several.

Declaration

public static double[, ] ConvertAmplitudeToPowerSpectrogram(double[, ] amplitudeM, double windowPower, int sampleRate)

Parameters

Type	Name	Description
Double[,]	amplitudeM	The frequency scaled amplitude spectrogram.
Double	windowPower	Power of the FFT window.
Int32	sampleRate	The sample rate of the original recording.

Returns

Type	Description
Double[,]	The Power Spectrogram as matrix. Each spectrum is a matrix row.

| Improve this Doc View Source

ConvertLinearSpectrogramToOctaveFreqScale(Double[,], FrequencyScale)

Converts a spectrogram having linear freq scale to one having an Octave freq scale. TODO: SHOULD DEVELOP A SEPARATE UNIT TEST for this method.

Declaration

public static double[, ] ConvertLinearSpectrogramToOctaveFreqScale(double[, ] inputSpgram, FrequencyScale freqScale)

Parameters

Type	Name	Description
Double[,]	inputSpgram
FrequencyScale	freqScale

Returns

Type	Description
Double[,]

| Improve this Doc View Source

GetDataReductionScale(FrequencyScale)

This method assumes that the linear spectrum is derived from a 512 frame with sr = 22050. It is a split linear-octave scale. The linear part is from 0-2 kHz with reduction by factor of 8. The octave part is obtained by setting octave divisions or tone count = 5.

Declaration

public static FrequencyScale GetDataReductionScale(FrequencyScale scale)

Parameters

Type	Name	Description
FrequencyScale	scale

Returns

Type	Description
FrequencyScale	a frequency scale for spectral-data reduction purposes.

| Improve this Doc View Source

GetFractionalOctaveBands(Double, Double, Int32)

Declaration

public static double[] GetFractionalOctaveBands(double minFreq, double maxFreq, int octaveDivisions)

Parameters

Type	Name	Description
Double	minFreq
Double	maxFreq
Int32	octaveDivisions

Returns

Type	Description
Double[]

| Improve this Doc View Source

GetFractionalOctaveBands(Double, Int32)

Returns an array of tones in one octave. The units are frequency in Hertz. NOTE: The octave is divided geometrically.

Declaration

public static double[] GetFractionalOctaveBands(double lowerBound, int subbandCount)

Parameters

Type	Name	Description
Double	lowerBound	The lower frquency bound of the octave.
Int32	subbandCount	The number of tones or frequency bins in the octave.

Returns

Type	Description
Double[]	The frequency of each tone in the octave.

| Improve this Doc View Source

GetGridLineLocations(Int32, Int32, Int32[,])

Declaration

public static int[, ] GetGridLineLocations(int nyquist, int linearBound, int[, ] octaveBinBounds)

Parameters

Type	Name	Description
Int32	nyquist
Int32	linearBound
Int32[,]	octaveBinBounds

Returns

Type	Description
Int32[,]

| Improve this Doc View Source

GetStandardOctaveScale(FrequencyScale)

Calculates the parameters for a standard mixed linear-octave frequency scale. IMPORTANT: It assumes that the nyquist, frame size and linear bound have already been set to valid values. What makes this scale "standard" is that the number of octaveDivsions/tones (T) is set equal to number of linear bins. The remainder of the spectrum will be reduced over T-tone octaves. Sensible values for linearUpperBound are 125, 250, 500, 1000. Note that when linearUpperBound = 500, the resulting spectrogram is very similar to the default MelScale. When nyquist=11025 and frameSize = 512, the default MelScale has 64 frequency bins and Linear500-octave has 66 frequency bands.

Declaration

public static FrequencyScale GetStandardOctaveScale(FrequencyScale scale)

Parameters

Type	Name	Description
FrequencyScale	scale

Returns

Type	Description
FrequencyScale

| Improve this Doc View Source

LinearToFullOctaveScale(Int32, Int32, Int32, Int32, Int32, Int32)

Returns the index bounds for a full octave scale - from lowest freq set by user to top freq.

Declaration

public static int[, ] LinearToFullOctaveScale(int sr, int frameSize, int finalBinCount, int lowerFreqBound, int upperFreqBound, int octaveDivisions)

Parameters

Type	Name	Description
Int32	sr	Sample rate of the source recording.
Int32	frameSize	Frame size of the source recording.
Int32	finalBinCount	Final Bin Count.
Int32	lowerFreqBound	Lower bound of the octave part of the final scale.
Int32	upperFreqBound	Upper bound of the octave scale, most likely the Nyquist.
Int32	octaveDivisions	Number of tones/divisions per octave.

Returns

Type	Description
Int32[,]

| Improve this Doc View Source

LinearToSplitLinearOctaveScale(Int32, Int32, Int32, Int32, Int32)

Returns a matrix that is used to transform a spectrum having linear Hz scale into a spectrum having an octave freq scale. The returned matrix is size N x 2, where N = the length of the output spectrum. In fact the op spectrum has a split Herz scale - bottom part linear, top part octave scaled. Column 0 of the returned matrix contains the index into linear spectrum. Column 1 of the returned matrix contains the Hertz value of the corresponding index into the linear spectrum.

Declaration

public static int[, ] LinearToSplitLinearOctaveScale(int nyquist, int frameSize, int lowerFreqBound, int upperFreqBound, int octaveDivisions)

Parameters

Type	Name	Description
Int32	nyquist
Int32	frameSize
Int32	lowerFreqBound
Int32	upperFreqBound
Int32	octaveDivisions

Returns

Type	Description
Int32[,]