/// <summary>
/// Computes a Scms model from the MFCC representation of a song.
/// </summary>
/// <param name="mfcc">Comirva.Audio.Util.Maths.Matrix mfcc</param>
/// <returns></returns>
public static Scms GetScmsNoInverse(Comirva.Audio.Util.Maths.Matrix mfccs, string name)
{
DbgTimer t = new DbgTimer();
t.Start();
Comirva.Audio.Util.Maths.Matrix mean = mfccs.Mean(2);
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE)
{
if (Analyzer.DEBUG_OUTPUT_TEXT)
{
mean.WriteText(name + "_mean.txt");
}
mean.DrawMatrixGraph(name + "_mean.png");
}
#endif
// Covariance
Comirva.Audio.Util.Maths.Matrix covarMatrix = mfccs.Cov(mean);
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE)
{
if (Analyzer.DEBUG_OUTPUT_TEXT)
{
covarMatrix.WriteText(name + "_covariance.txt");
}
covarMatrix.DrawMatrixGraph(name + "_covariance.png");
}
#endif
Comirva.Audio.Util.Maths.Matrix covarMatrixInv = new Comirva.Audio.Util.Maths.Matrix(covarMatrix.Rows, covarMatrix.Columns);
// Store the Mean, Covariance, Inverse Covariance in an optimal format.
int dim = mean.Rows;
Scms s = new Scms(dim);
int l = 0;
for (int i = 0; i < dim; i++)
{
s.mean[i] = (float)mean.MatrixData[i][0];
for (int j = i; j < dim; j++)
{
s.cov[l] = (float)covarMatrix.MatrixData[i][j];
s.icov[l] = (float)covarMatrixInv.MatrixData[i][j];
l++;
}
}
Dbg.WriteLine("GetScmsNoInverse - Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return(s);
}
public AudioFeature Calculate(double[] input)
{
//pack the mfccs into a pointlist
double[][] mfccCoefficients = mfcc.Process(input);
//check if element 0 exists
if(mfccCoefficients.Length == 0)
throw new ArgumentException("The input stream is to short to process;");
//create mfcc matrix
Matrix mfccs = new Matrix(mfccCoefficients);
#if DEBUG
mfccs.WriteText("mfccdata-mandelellis.txt");
mfccs.DrawMatrixGraph("mfccdata-mandelellis.png");
#endif
// compute mean
//Matrix mean = mfccs.Mean(1).Transpose();
Matrix mean = mfccs.Mean(2);
#if DEBUG
mean.WriteText("mean-mandelellis.txt");
mean.DrawMatrixGraph("mean-mandelellis.png");
#endif
// create covariance matrix
Matrix covarMatrix = mfccs.Cov();
#if DEBUG
covarMatrix.WriteText("covariance-mandelellis.txt");
covarMatrix.DrawMatrixGraph("covariance-mandelellis.png");
#endif
// Inverse Covariance
Matrix covarMatrixInv;
try {
//covarMatrixInv = covarMatrix.Inverse();
covarMatrixInv = covarMatrix.InverseGausJordan();
} catch (Exception) {
Console.Error.WriteLine("Mandel Ellis Extraction Failed!");
return null;
}
#if DEBUG
covarMatrixInv.WriteText("inverse_covariance-mandelellis.txt");
covarMatrixInv.DrawMatrixGraph("inverse_covariance-mandelellis.png");
#endif
MandelEllis.GmmMe gmmMe = new MandelEllis.GmmMe(mean, covarMatrix, covarMatrixInv);
MandelEllis mandelEllis = new MandelEllis(gmmMe);
return mandelEllis;
}
/// <summary>
/// Init the MFCC class
/// </summary>
/// <param name="sampleRate">float sampes per second, must be greater than zero; not
/// whole-numbered values get rounded</param>
/// <param name="windowSize">int size of window; must be 2^n and at least 32</param>
/// <param name="numberCoefficients">int must be grate or equal to 1 and smaller than
/// the number of filters</param>
/// <param name="useFirstCoefficient">boolean indicates whether the first coefficient
/// of the dct process should be used in the
/// mfcc feature vector or not</param>
/// <param name="minFreq">double start of the interval to place the mel-filters in</param>
/// <param name="maxFreq">double end of the interval to place the mel-filters in</param>
/// <param name="numberFilters">int number of mel-filters to place in the interval</param>
private void Initialize(float sampleRate, int windowSize, int numberCoefficients, bool useFirstCoefficient, double minFreq, double maxFreq, int numberFilters)
{
//check for correct window size
if(windowSize < 32)
{
throw new Exception("window size must be at least 32");
}
else
{
int i = 32;
while(i < windowSize && i < Int32.MaxValue)
i = i << 1;
if(i != windowSize)
throw new Exception("window size must be 2^n");
}
//check sample rate
sampleRate = (float)Math.Round(sampleRate);
if(sampleRate < 1)
throw new Exception("sample rate must be at least 1");
//check numberFilters
if(numberFilters < 2 || numberFilters > (windowSize/2) + 1)
throw new Exception("number filters must be at least 2 and smaller than the nyquist frequency");
//check numberCoefficients
if(numberCoefficients < 1 || numberCoefficients >= numberFilters)
throw new Exception("the number of coefficients must be greater or equal to 1 and samller than the number of filters");
//check minFreq/maxFreq
if(minFreq <= 0 || minFreq > maxFreq || maxFreq > 88200.0f)
throw new Exception("the min. frequency must be greater 0 smaller than the max. frequency, which must be smaller than 88200.0");;
this.sampleRate = sampleRate;
this.windowSize = windowSize;
this.hopSize = windowSize/2; //50% Overleap
this.baseFreq = sampleRate/windowSize;
this.numberCoefficients = numberCoefficients;
this.useFirstCoefficient = useFirstCoefficient;
this.minFreq = minFreq;
this.maxFreq = maxFreq;
this.numberFilters = numberFilters;
//create buffers
inputData = new double[windowSize];
buffer = new double[windowSize];
//store filter weights and DCT matrix due to performance reason
melFilterBanks = GetMelFilterBanks();
#if DEBUG
if (Mirage.Analyzer.DEBUG_INFO_VERBOSE) {
melFilterBanks.DrawMatrixGraph("melfilters-comirva.png");
}
#endif
dctMatrix = GetDCTMatrix();
#if DEBUG
if (Mirage.Analyzer.DEBUG_INFO_VERBOSE) {
dctMatrix.DrawMatrixGraph("dct-comirva.png");
}
#endif
//create power fft object
normalizedPowerFFT = new FFT(FFT.FFT_NORMALIZED_POWER, windowSize, FFT.WND_HANNING);
}
/// <summary>
/// Computes a Scms model from the MFCC representation of a song.
/// </summary>
/// <param name="mfcc">Comirva.Audio.Util.Maths.Matrix mfcc</param>
/// <returns></returns>
public static Scms GetScms(Comirva.Audio.Util.Maths.Matrix mfccs, string name)
{
DbgTimer t = new DbgTimer();
t.Start();
Comirva.Audio.Util.Maths.Matrix mean = mfccs.Mean(2);
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE)
{
if (Analyzer.DEBUG_OUTPUT_TEXT)
{
mean.WriteText(name + "_mean.txt");
}
mean.DrawMatrixGraph(name + "_mean.png");
}
#endif
// Covariance
Comirva.Audio.Util.Maths.Matrix covarMatrix = mfccs.Cov(mean);
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE)
{
if (Analyzer.DEBUG_OUTPUT_TEXT)
{
covarMatrix.WriteText(name + "_covariance.txt");
}
covarMatrix.DrawMatrixGraph(name + "_covariance.png");
}
#endif
// Inverse Covariance
Comirva.Audio.Util.Maths.Matrix covarMatrixInv;
try {
covarMatrixInv = covarMatrix.InverseGausJordan();
} catch (Exception) {
Dbg.WriteLine("MatrixSingularException - Scms failed!");
return(null);
}
#if DEBUG
if (Analyzer.DEBUG_INFO_VERBOSE)
{
if (Analyzer.DEBUG_OUTPUT_TEXT)
{
covarMatrixInv.WriteAscii(name + "_inverse_covariance.ascii");
}
covarMatrixInv.DrawMatrixGraph(name + "_inverse_covariance.png");
}
#endif
// Store the Mean, Covariance, Inverse Covariance in an optimal format.
int dim = mean.Rows;
Scms s = new Scms(dim);
int l = 0;
for (int i = 0; i < dim; i++)
{
s.mean[i] = (float)mean.MatrixData[i][0];
for (int j = i; j < dim; j++)
{
s.cov[l] = (float)covarMatrix.MatrixData[i][j];
s.icov[l] = (float)covarMatrixInv.MatrixData[i][j];
l++;
}
}
Dbg.WriteLine("Compute Scms - Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return(s);
}