public void ComputeInverseMatrixUsingLomontTableFFT(Matrix m, int column, ref double[] signal, int winsize, int overlap)
{
double[] spectrogramWindow = m.GetColumn(column);
// extend window with the inverse duplicate array
int len = spectrogramWindow.Length;
var extendedWindow = new double[len * 2];
Array.Copy(spectrogramWindow, extendedWindow, len);
for (int i = 1; i < len; i++)
{
extendedWindow[len + i] = spectrogramWindow[len - i];
}
double[] complexSignal = FFTUtils.DoubleToComplexDouble(extendedWindow);
lomonFFT.TableFFT(complexSignal, false);
double[] window = win.Window;
// multiply by window w/ overlap-add
int N = complexSignal.Length / 2;
var returnArray = new double[N];
for (int j = 0; j < N; j++)
{
double re = complexSignal[2 * j] / Math.Sqrt(winsize);
//double img = complexSignal[2*j + 1];
returnArray[j] = re * window[j]; // smooth yet another time (also did this when doing FFT)
// overlap-add method
// scale with 2 just because the volume got so much lower when using a second smoothing filter when reconstrcting
signal[j + overlap * column] = signal[j + overlap * column] + returnArray[j]; // * 3;
}
}
/// <summary>
/// Generate a spectrogram array spaced linearily
/// </summary>
/// <param name="samples">audio data</param>
/// <param name="fftWindowsSize">fft window size</param>
/// <param name="fftOverlap">overlap in number of samples (normaly half of the fft window size) [low number = high overlap]</param>
/// <returns>spectrogram jagged array</returns>
public static double[][] CreateSpectrogramFFTWLIB_INPLACE(float[] samples, int fftWindowsSize, int fftOverlap)
{
int numberOfSamples = samples.Length;
// overlap must be an integer smaller than the window size
// half the windows size is quite normal
double[] windowArray = FFTWindow.GetWindowFunction(FFTWindowType.HANNING, fftWindowsSize);
// width of the segment - e.g. split the file into 78 time slots (numberOfSegments) and do analysis on each slot
int numberOfSegments = (numberOfSamples - fftWindowsSize) / fftOverlap;
var frames = new double[numberOfSegments][];
var signal = new double[fftWindowsSize];
for (int i = 0; i < numberOfSegments; i++)
{
// apply Hanning Window
for (int j = 0; j < fftWindowsSize; j++)
{
// Weight by Hann Window
signal[j] = (double)(windowArray[j] * samples[i * fftOverlap + j]);
}
// perform the FFT
FFTW_FFT_R2R(ref signal, ref signal, fftWindowsSize, FFTMethod.DFT);
// get the result
double[] complexDout = FFTUtils.HC2C(signal);
frames[i] = Abs(complexDout);
}
return(frames);
}
public void ComputeMatrixUsingLomontTableFFT(ref Matrix m, int column, float[] audiodata, int pos)
{
// apply the window method (e.g HammingWindow, HannWindow etc)
win.Apply(ref data, audiodata, pos);
double[] complexSignal = FFTUtils.FloatToComplexDouble(data);
lomonFFT.TableFFT(complexSignal, true);
int row = 0;
for (int i = 0; i < complexSignal.Length / 4; i += 2)
{
double re = complexSignal[2 * i];
double img = complexSignal[2 * i + 1];
m.MatrixData[row][column] = Math.Sqrt((re * re + img * img) * complexSignal.Length / 2);
row++;
}
}
static double[] FFTWLIB_INPLACE(double[] signal)
{
int N = signal.Length;
double[] din = signal;
var dout = new double[N];
// perform the FFT
FFTUtils.FFTW_FFT_R2R(ref din, ref dout, N, FFTUtils.FFTMethod.DFT);
// get the result
double[] complexDout = FFTUtils.HC2C(dout);
var spectrum_fft_abs = FFTUtils.Abs(complexDout);
//Export.ExportCSV("audio_buffer_padded.csv", din);
//Export.ExportCSV("spectrum_fft_abs.csv", spectrum_fft_abs, fftSize);
return(spectrum_fft_abs);
}
/// <summary>
/// Generate a spectrogram array spaced linearily
/// </summary>
/// <param name="samples">audio data</param>
/// <param name="fftWindowsSize">fft window size</param>
/// <param name="fftOverlap">overlap in number of samples (normaly half of the fft window size) [low number = high overlap]</param>
/// <returns>spectrogram jagged array</returns>
public static double[][] CreateSpectrogramFFTW(float[] samples, int fftWindowsSize, int fftOverlap)
{
int numberOfSamples = samples.Length;
// overlap must be an integer smaller than the window size
// half the windows size is quite normal
double[] windowArray = FFTWindow.GetWindowFunction(FFTWindowType.HANNING, fftWindowsSize);
// width of the segment - e.g. split the file into 78 time slots (numberOfSegments) and do analysis on each slot
int numberOfSegments = (numberOfSamples - fftWindowsSize) / fftOverlap;
var frames = new double[numberOfSegments][];
// even - Re, odd - Img
var realSignal = new double[fftWindowsSize];
double lengthSqrt = Math.Sqrt(fftWindowsSize);
for (int i = 0; i < numberOfSegments; i++)
{
// apply Hanning Window
for (int j = 0; j < fftWindowsSize; j++)
{
// Weight by Hann Window
realSignal[j] = (double)(windowArray[j] * samples[i * fftOverlap + j]);
}
// FFT transform for gathering the spectrum
var complexOutput = FFTWPaddedFFT(ref realSignal);
// 4096 real numbers on input processed by FFTW dft_r2c_1d transform gives
// 4096/2+1 = 2049 complex numbers at output
// remove the value as FFTW returns one extra complex value not needed
var complexFixed = new double[complexOutput.Length - 1];
Array.Copy(complexOutput, complexFixed, complexFixed.Length);
// get the result
var spectrum_fft_abs = FFTUtils.Abs(complexFixed);
frames[i] = spectrum_fft_abs;
}
return(frames);
}
// seem to the be the fastest FFT?
static double[] FFTWLIB(double[] signal)
{
var complexSignal = FFTUtils.DoubleToComplexDouble(signal);
// prepare the input arrays
var complexInput = new fftw_complexarray(complexSignal);
var complexOutput = new fftw_complexarray(complexSignal.Length / 2);
fftw_plan fft = fftw_plan.dft_1d(complexSignal.Length / 2, complexInput, complexOutput, fftw_direction.Forward, fftw_flags.Estimate);
// perform the FFT
fft.Execute();
// get the result
var spectrum_fft_abs = complexOutput.Abs;
//Export.ExportCSV("audio_buffer_padded2.csv", signal);
//Export.ExportCSV("spectrum_fft_abs2.csv", spectrum_fft_abs2, fftSize);
// free up memory
complexInput = null;
complexOutput = null;
return(spectrum_fft_abs);
}
