//##################################################################################################################################
/// <summary>
/// NOTE!!!! The decibel array has been normalised in 0 - 1.
/// </summary>
protected static Tuple <double[, ], double[]> MakeCepstrogram(SonogramConfig config, double[,] matrix, double[] decibels, int sampleRate)
{
double[,] m = matrix;
int nyquist = sampleRate / 2;
double epsilon = config.epsilon;
bool includeDelta = config.mfccConfig.IncludeDelta;
bool includeDoubleDelta = config.mfccConfig.IncludeDoubleDelta;
//Log.WriteIfVerbose(" MakeCepstrogram(matrix, decibels, includeDelta=" + includeDelta + ", includeDoubleDelta=" + includeDoubleDelta + ")");
//(i) APPLY FILTER BANK
int bandCount = config.mfccConfig.FilterbankCount;
bool doMelScale = config.mfccConfig.DoMelScale;
int ccCount = config.mfccConfig.CcCount;
int fftBinCount = config.FreqBinCount; //number of Hz bands = 2^N +1. Subtract DC bin
int minHz = config.MinFreqBand ?? 0;
int maxHz = config.MaxFreqBand ?? nyquist;
Log.WriteIfVerbose("ApplyFilterBank(): Dim prior to filter bank =" + matrix.GetLength(1));
//error check that filterBankCount < FFTbins
if (bandCount > fftBinCount)
{
throw new Exception(
"## FATAL ERROR in BaseSonogram.MakeCepstrogram():- Can't calculate cepstral coeff. FilterbankCount > FFTbins. (" +
bandCount + " > " + fftBinCount + ")\n\n");
}
//this is the filter count for full bandwidth 0-Nyquist. This number is trimmed proportionately to fit the required bandwidth.
if (doMelScale)
{
m = MFCCStuff.MelFilterBank(m, bandCount, nyquist, minHz, maxHz); // using the Greg integral
}
else
{
m = MFCCStuff.LinearFilterBank(m, bandCount, nyquist, minHz, maxHz);
}
Log.WriteIfVerbose("\tDim after filter bank=" + m.GetLength(1) + " (Max filter bank=" + bandCount + ")");
//(ii) CONVERT AMPLITUDES TO DECIBELS
m = MFCCStuff.DecibelSpectra(m, config.WindowPower, sampleRate, epsilon); //from spectrogram
//(iii) NOISE REDUCTION
var tuple1 = SNR.NoiseReduce(m, config.NoiseReductionType, config.NoiseReductionParameter);
m = tuple1.Item1;
//(iv) calculate cepstral coefficients
m = MFCCStuff.Cepstra(m, ccCount);
//(v) NormaliseMatrixValues
m = DataTools.normalise(m);
//(vi) Calculate the full range of MFCC coefficients ie including decibel and deltas, etc
m = MFCCStuff.AcousticVectors(m, decibels, includeDelta, includeDoubleDelta);
var tuple2 = Tuple.Create(m, tuple1.Item2);
return(tuple2); // return matrix and full bandwidth modal noise profile
}
/// <summary>
/// The data passed to this method must be the Spectral sonogram.
/// </summary>
public static Tuple <double[, ], double[]> GetCepstrogram(double[,] data, int minHz, int maxHz,
int freqBinCount, double freqBinWidth, bool doMelScale, int ccCount)
{
ImageTools.DrawMatrix(data, @"C:\SensorNetworks\Output\MFCC_LewinsRail\tempImage1.jpg", false);
double[,] m = SpectrogramTools.ExtractFreqSubband(data, minHz, maxHz, doMelScale, freqBinCount, freqBinWidth);
ImageTools.DrawMatrix(m, @"C:\SensorNetworks\Output\MFCC_LewinsRail\tempImage2.jpg", false);
//DO NOT DO NOISE REDUCTION BECAUSE ALREADY DONE
//double[] modalNoise = SNR.CalculateModalNoise(m, 7); //calculate modal noise profile and smooth
//m = SNR.NoiseReduce_Standard(m, modalNoise);
//m = SNR.NoiseReduce_FixedRange(m, this.Configuration.DynamicRange);
m = MFCCStuff.Cepstra(m, ccCount);
m = DataTools.normalise(m);
ImageTools.DrawMatrix(m, @"C:\SensorNetworks\Output\MFCC_LewinsRail\tempImage3.jpg", false);
return(Tuple.Create(m, (double[])null));
}
public static Tuple <double[]> Execute_MFCC_XCOR(double[,] target, double dynamicRange, SpectrogramStandard sonogram,
List <AcousticEvent> segments, int minHz, int maxHz, double minDuration)
{
Log.WriteLine("SEARCHING FOR EVENTS LIKE TARGET.");
if (segments == null)
{
return(null);
}
int minBin = (int)(minHz / sonogram.FBinWidth);
int maxBin = (int)(maxHz / sonogram.FBinWidth);
int targetLength = target.GetLength(0);
//set up the matrix of cosine coefficients
int coeffCount = 12; //only use first 12 coefficients.
int binCount = target.GetLength(1); //number of filters in filter bank
double[,] cosines = MFCCStuff.Cosines(binCount, coeffCount + 1); //set up the cosine coefficients
//adjust target's dynamic range to that set by user
target = SNR.SetDynamicRange(target, 3.0, dynamicRange); //set event's dynamic range
target = MFCCStuff.Cepstra(target, coeffCount, cosines);
double[] v1 = DataTools.Matrix2Array(target);
v1 = DataTools.normalise2UnitLength(v1);
string imagePath2 = @"C:\SensorNetworks\Output\FELT_Currawong\target.png";
var result1 = BaseSonogram.Data2ImageData(target);
var image = result1.Item1;
ImageTools.DrawMatrix(image, 1, 1, imagePath2);
double[] scores = new double[sonogram.FrameCount];
foreach (AcousticEvent av in segments)
{
Log.WriteLine("SEARCHING SEGMENT.");
int startRow = (int)Math.Round(av.TimeStart * sonogram.FramesPerSecond);
int endRow = (int)Math.Round(av.TimeEnd * sonogram.FramesPerSecond);
if (endRow >= sonogram.FrameCount)
{
endRow = sonogram.FrameCount - 1;
}
endRow -= targetLength;
if (endRow <= startRow)
{
endRow = startRow + 1; //want minimum of one row
}
for (int r = startRow; r < endRow; r++)
{
double[,] matrix = DataTools.Submatrix(sonogram.Data, r, minBin, r + targetLength - 1, maxBin);
matrix = SNR.SetDynamicRange(matrix, 3.0, dynamicRange); //set event's dynamic range
//string imagePath2 = @"C:\SensorNetworks\Output\FELT_Gecko\compare.png";
//var image = BaseSonogram.Data2ImageData(matrix);
//ImageTools.DrawMatrix(image, 1, 1, imagePath2);
matrix = MFCCStuff.Cepstra(matrix, coeffCount, cosines);
double[] v2 = DataTools.Matrix2Array(matrix);
v2 = DataTools.normalise2UnitLength(v2);
double crossCor = DataTools.DotProduct(v1, v2);
scores[r] = crossCor;
} //end of rows in segment
} //foreach (AcousticEvent av in segments)
var tuple = Tuple.Create(scores);
return(tuple);
}
