public float[][] CreateLogSpectrogram(
float[] samples, IWindowFunction windowFunction, AudioServiceConfiguration configuration)
{
NormalizeInPlace(samples);
int width = (samples.Length - configuration.WdftSize) / configuration.Overlap; /*width of the image*/
float[][] frames = new float[width][];
float[] complexSignal = new float[2 * configuration.WdftSize]; /*even - Re, odd - Img*/
double[] window = windowFunction.GetWindow(configuration.WdftSize);
int[] logFrequenciesIndexes = GenerateLogFrequencies(configuration);
for (int i = 0; i < width; i++)
{
// take 371 ms each 11.6 ms (2048 samples each 64 samples)
for (int j = 0; j < configuration.WdftSize /*2048*/; j++)
{
complexSignal[(2 * j)] = (float)(window[j] * samples[(i * configuration.Overlap) + j]);
/*Weight by Hann Window*/
complexSignal[(2 * j) + 1] = 0;
}
// FFT transform for gathering the spectrum
Fourier.FFT(complexSignal, configuration.WdftSize, FourierDirection.Forward);
frames[i] = ExtractLogBins(complexSignal, logFrequenciesIndexes, configuration.LogBins);
}
return frames;
}
public void ComputeInverseComirvaMatrixUsingLomontRealFFT(Comirva.Audio.Util.Maths.Matrix m, int column, ref double[] signal, int winsize, int hopsize)
{
double[] spectrogramWindow = m.GetColumn(column);
// extend window with the inverse duplicate array
int len = spectrogramWindow.Length;
double[] extendedWindow = new double[len * 2];
Array.Copy(spectrogramWindow, extendedWindow, len);
for (int i = 1; i < len; i++)
{
extendedWindow[len + i] = spectrogramWindow[len - i];
}
// ifft input must contain the FFT values
// r0, r(n/2), r1, i1, r2, i2 ...
// Perform the ifft and take just the real part
double[] ifft = new double[winsize * 2];
ifft[0] = extendedWindow[0];
ifft[1] = extendedWindow[winsize / 2];
for (int i = 1; i < extendedWindow.Length; i++)
{
ifft[2 * i] = extendedWindow[i];
}
lomonFFT.RealFFT(ifft, false);
double[] window = win.GetWindow();
// multiply by window w/ overlap-add
int N = ifft.Length / 2;
double[] returnArray = new double[N];
for (int j = 0; j < N; j++)
{
double re = ifft[2 * j] / Math.Sqrt(winsize);
returnArray[j] = re * window[j]; // smooth yet another time (also did this when doing FFT)
// overlap-add method
// scale with 5 just because the volume got so much lower when using a second smoothing filter when reconstrcting
signal[j + hopsize * column] = signal[j + hopsize * column] + returnArray[j] * 5;
}
}
public double[][] CreateSpectrogram(string pathToFilename, IWindowFunction windowFunction, int sampleRate, int overlap, int wdftSize)
{
// read 5512 Hz, Mono, PCM, with a specific proxy
float[] samples = ReadMonoFromFile(pathToFilename, sampleRate, 0, 0);
NormalizeInPlace(samples);
int width = (samples.Length - wdftSize) / overlap; /*width of the image*/
double[][] frames = new double[width][];
double[] complexSignal = new double[2 * wdftSize]; /*even - Re, odd - Img, thats how Exocortex works*/
double[] window = windowFunction.GetWindow();
for (int i = 0; i < width; i++)
{
// take 371 ms each 11.6 ms (2048 samples each 64 samples)
for (int j = 0; j < wdftSize; j++)
{
// Weight by Hann Window
complexSignal[2 * j] = window[j] * samples[(i * overlap) + j];
// need to clear out as fft modifies buffer (phase)
complexSignal[(2 * j) + 1] = 0;
}
lomonFFT.TableFFT(complexSignal, true);
// When the input is purely real, its transform is Hermitian,
// i.e., the component at frequency f_k is the complex conjugate of the component
// at frequency -f_k, which means that for real inputs there is no information
// in the negative frequency components that is not already available from the
// positive frequency components.
// Thus, n input points produce n/2+1 complex output points.
// The inverses of this family assumes the same symmetry of its input,
// and for an output of n points uses n/2+1 input points.
// Transform output contains, for a transform of size N,
// N/2+1 complex numbers, i.e. 2*(N/2+1) real numbers
// our transform is of size N+1, because the histogram has n+1 bins
double[] band = new double[(wdftSize / 2)]; // Don't add te last band, i.e. + 1 is removed
for (int j = 0; j < (wdftSize / 2); j++) // Don't add te last band, i.e. + 1 is removed
{
double re = complexSignal[2 * j];
double img = complexSignal[(2 * j) + 1];
band[j] = Math.Sqrt(((re * re) + (img * img)) * wdftSize);
}
frames[i] = band;
}
return(frames);
}
public double[][] CreateSpectrogram(string pathToFilename, IWindowFunction windowFunction, int sampleRate, int overlap, int wdftSize)
{
// read 5512 Hz, Mono, PCM, with a specific proxy
float[] samples = ReadMonoFromFile(pathToFilename, sampleRate, 0, 0);
NormalizeInPlace(samples);
int width = (samples.Length - wdftSize) / overlap; /*width of the image*/
double[][] frames = new double[width][];
double[] complexSignal = new double[2 * wdftSize]; /*even - Re, odd - Img, thats how Exocortex works*/
double[] window = windowFunction.GetWindow();
for (int i = 0; i < width; i++)
{
// take 371 ms each 11.6 ms (2048 samples each 64 samples)
for (int j = 0; j < wdftSize; j++)
{
// Weight by Hann Window
complexSignal[2 * j] = window[j] * samples[(i * overlap) + j];
// need to clear out as fft modifies buffer (phase)
complexSignal[(2 * j) + 1] = 0;
}
lomonFFT.TableFFT(complexSignal, true);
// When the input is purely real, its transform is Hermitian,
// i.e., the component at frequency f_k is the complex conjugate of the component
// at frequency -f_k, which means that for real inputs there is no information
// in the negative frequency components that is not already available from the
// positive frequency components.
// Thus, n input points produce n/2+1 complex output points.
// The inverses of this family assumes the same symmetry of its input,
// and for an output of n points uses n/2+1 input points.
// Transform output contains, for a transform of size N,
// N/2+1 complex numbers, i.e. 2*(N/2+1) real numbers
// our transform is of size N+1, because the histogram has n+1 bins
double[] band = new double[(wdftSize / 2)]; // Don't add te last band, i.e. + 1 is removed
for (int j = 0; j < (wdftSize / 2); j++) // Don't add te last band, i.e. + 1 is removed
{
double re = complexSignal[2 * j];
double img = complexSignal[(2 * j) + 1];
band[j] = Math.Sqrt( ((re * re) + (img * img)) * wdftSize);
}
frames[i] = band;
}
return frames;
}
public double[][] CreateLogSpectrogram(
float[] samples, IWindowFunction windowFunction, AudioServiceConfiguration configuration)
{
DbgTimer t = new DbgTimer();
t.Start();
if (configuration.NormalizeSignal)
{
NormalizeInPlace(samples);
}
int width = (samples.Length - configuration.WindowSize) / configuration.Overlap; /*width of the image*/
double[][] frames = new double[width][];
int[] logFrequenciesIndexes = GenerateLogFrequencies(configuration);
double[] window = windowFunction.GetWindow();
for (int i = 0; i < width; i++)
{
double[] complexSignal = new double[2 * configuration.WindowSize]; /*even - Re, odd - Img, thats how Exocortex works*/
// take 371 ms each 11.6 ms (2048 samples each 64 samples, samplerate 5512)
// or 256 ms each 16 ms (8192 samples each 512 samples, samplerate 32000)
for (int j = 0; j < configuration.WindowSize; j++)
{
// Weight by Hann Window
complexSignal[2 * j] = window[j] * samples[(i * configuration.Overlap) + j];
// need to clear out as fft modifies buffer (phase)
complexSignal[(2 * j) + 1] = 0;
}
lomonFFT.TableFFT(complexSignal, true);
frames[i] = ExtractLogBins(complexSignal, logFrequenciesIndexes, configuration.LogBins);
}
Dbg.WriteLine("Create Log Spectrogram - Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return(frames);
}
public double[][] CreateLogSpectrogram(
float[] samples, IWindowFunction windowFunction, AudioServiceConfiguration configuration)
{
// Explode samples to the range of 16 bit shorts (–32,768 to 32,767)
// Matlab multiplies with 2^15 (32768)
// e.g. if( max(abs(speech))<=1 ), speech = speech * 2^15; end;
//MathUtils.Multiply(ref samples, Analyzer.AUDIO_MULTIPLIER); // 65536
if (configuration.NormalizeSignal)
{
NormalizeInPlace(samples);
}
int width = (samples.Length - configuration.WdftSize) / configuration.Overlap; /*width of the image*/
double[][] frames = new double[width][];
int[] logFrequenciesIndexes = GenerateLogFrequencies(configuration);
//double[] window = windowFunction.GetWindow(configuration.WdftSize);
double[] window = windowFunction.GetWindow();
for (int i = 0; i < width; i++)
{
double[] complexSignal = new double[2 * configuration.WdftSize]; /*even - Re, odd - Img, thats how Exocortex works*/
// take 371 ms each 11.6 ms (2048 samples each 64 samples)
for (int j = 0; j < configuration.WdftSize; j++)
{
// Weight by Hann Window
complexSignal[2 * j] = window[j] * samples[(i * configuration.Overlap) + j];
// need to clear out as fft modifies buffer (phase)
complexSignal[(2 * j) + 1] = 0;
}
lomonFFT.TableFFT(complexSignal, true);
frames[i] = ExtractLogBins(complexSignal, logFrequenciesIndexes, configuration.LogBins);
}
return frames;
}
public float[][] CreateSpectrogram(string pathToFilename, IWindowFunction windowFunction, int sampleRate, int overlap, int wdftSize)
{
// read 5512 Hz, Mono, PCM, with a specific proxy
float[] samples = ReadMonoFromFile(pathToFilename, sampleRate, 0, 0);
NormalizeInPlace(samples);
int width = (samples.Length - wdftSize) / overlap; /*width of the image*/
float[][] frames = new float[width][];
float[] complexSignal = new float[2 * wdftSize]; /*even - Re, odd - Img*/
double[] window = windowFunction.GetWindow(wdftSize);
for (int i = 0; i < width; i++)
{
// take 371 ms each 11.6 ms (2048 samples each 64 samples)
for (int j = 0; j < wdftSize; j++)
{
complexSignal[2 * j] = (float)(window[j] * samples[(i * overlap) + j]);
/*Weight by Hann Window*/
complexSignal[(2 * j) + 1] = 0;
}
Fourier.FFT(complexSignal, wdftSize, FourierDirection.Forward);
float[] band = new float[(wdftSize / 2) + 1];
for (int j = 0; j < (wdftSize / 2) + 1; j++)
{
double re = complexSignal[2 * j];
double img = complexSignal[(2 * j) + 1];
re /= (float)wdftSize / 2;
img /= (float)wdftSize / 2;
band[j] = (float)Math.Sqrt((re * re) + (img * img));
}
frames[i] = band;
}
return frames;
}
public double[][] CreateLogSpectrogram(
float[] samples, IWindowFunction windowFunction, AudioServiceConfiguration configuration)
{
DbgTimer t = new DbgTimer();
t.Start ();
if (configuration.NormalizeSignal)
{
NormalizeInPlace(samples);
}
int width = (samples.Length - configuration.WindowSize) / configuration.Overlap; /*width of the image*/
double[][] frames = new double[width][];
int[] logFrequenciesIndexes = GenerateLogFrequencies(configuration);
double[] window = windowFunction.GetWindow();
for (int i = 0; i < width; i++)
{
double[] complexSignal = new double[2 * configuration.WindowSize]; /*even - Re, odd - Img, thats how Exocortex works*/
// take 371 ms each 11.6 ms (2048 samples each 64 samples, samplerate 5512)
// or 256 ms each 16 ms (8192 samples each 512 samples, samplerate 32000)
for (int j = 0; j < configuration.WindowSize; j++)
{
// Weight by Hann Window
complexSignal[2 * j] = window[j] * samples[(i * configuration.Overlap) + j];
// need to clear out as fft modifies buffer (phase)
complexSignal[(2 * j) + 1] = 0;
}
lomonFFT.TableFFT(complexSignal, true);
frames[i] = ExtractLogBins(complexSignal, logFrequenciesIndexes, configuration.LogBins);
}
Dbg.WriteLine ("Create Log Spectrogram - Execution Time: {0} ms", t.Stop().TotalMilliseconds);
return frames;
}