public List <SpectralImage> CreateLogSpectrogram(AudioSamples audioSamples, SpectrogramConfig configuration)
{
int wdftSize = configuration.WdftSize;
int width = (audioSamples.Samples.Length - wdftSize) / configuration.Overlap;
if (width < 1)
{
return(new List <SpectralImage>());
}
float[] frames = new float[width * configuration.LogBins];
ushort[] logFrequenciesIndexes = logUtility.GenerateLogFrequenciesRanges(audioSamples.SampleRate, configuration);
float[] window = configuration.Window.GetWindow(wdftSize);
float[] samples = audioSamples.Samples;
unsafe
{
Parallel.For(0, width, index =>
{
float *fftArray = stackalloc float[wdftSize];
CopyAndWindow(fftArray, samples, index * configuration.Overlap, window);
fftServiceUnsafe.FFTForwardInPlace(fftArray, wdftSize);
ExtractLogBins(fftArray, logFrequenciesIndexes, configuration.LogBins, wdftSize, frames, index);
});
}
var images = CutLogarithmizedSpectrum(frames, audioSamples.SampleRate, configuration);
ScaleFullSpectrum(images, configuration);
return(images);
}
public List <SpectralImage> CreateLogSpectrogram(AudioSamples audioSamples, SpectrogramConfig configuration)
{
int width = (audioSamples.Samples.Length - configuration.WdftSize) / configuration.Overlap;
if (width < 1)
{
return(new List <SpectralImage>());
}
float[][] frames = new float[width][];
int[] logFrequenciesIndexes = logUtility.GenerateLogFrequenciesRanges(audioSamples.SampleRate, configuration);
for (int i = 0; i < width; i++)
{
float[] complexSignal = fftService.FFTForward(audioSamples.Samples, i * configuration.Overlap, configuration.WdftSize);
frames[i] = ExtractLogBins(complexSignal, logFrequenciesIndexes, configuration.LogBins);
}
return(CutLogarithmizedSpectrum(frames, audioSamples.SampleRate, configuration));
}
public float[][] CreateLogSpectrogram(float[] samples, IFingerprintConfiguration configuration)
{
if (configuration.NormalizeSignal)
{
audioSamplesNormalizer.NormalizeInPlace(samples);
}
int width = (samples.Length - configuration.WdftSize) / configuration.Overlap; /*width of the image*/
float[][] frames = new float[width][];
int[] logFrequenciesIndexes = logUtility.GenerateLogFrequenciesRanges(configuration);
for (int i = 0; i < width; i++)
{
float[] complexSignal = fftService.FFTForward(samples, i * configuration.Overlap, configuration.WdftSize);
frames[i] = ExtractLogBins(complexSignal, logFrequenciesIndexes, configuration.LogBins);
}
return(frames);
}
public void GenerateLogFrequenciesRangesTest()
{
var defaultConfig = new DefaultSpectrogramConfig {
UseDynamicLogBase = false, LogBase = 10
};
float[] logSpacedFrequencies = new[] // generated in matlab with logspace(2.50242712, 3.3010299957, 33)
{
318.00f, 336.81f, 356.73f, 377.83f, 400.18f, 423.85f, 448.92f, 475.47f, 503.59f, 533.38f, 564.92f,
598.34f, 633.73f, 671.21f, 710.91f, 752.96f, 797.50f, 844.67f, 894.63f, 947.54f, 1003.58f, 1062.94f,
1125.81f, 1192.40f, 1262.93f, 1337.63f, 1416.75f, 1500.54f, 1589.30f, 1683.30f, 1782.86f, 1888.31f,
2000f
};
int[] indexes = logUtility.GenerateLogFrequenciesRanges(defaultFingerprintConfiguration.SampleRate, defaultConfig);
for (int i = 0; i < logSpacedFrequencies.Length; i++)
{
var logSpacedFrequency = logSpacedFrequencies[i];
int index = logUtility.FrequencyToSpectrumIndex(logSpacedFrequency, defaultFingerprintConfiguration.SampleRate, defaultConfig.WdftSize);
Assert.AreEqual(index, indexes[i]);
}
}
public List <SpectralImage> CreateLogSpectrogram(AudioSamples audioSamples, SpectrogramConfig configuration)
{
using (new DebugTimer("CreateLogSpectrogram()"))
{
int wdftSize = configuration.WdftSize;
int width = (audioSamples.Samples.Length - wdftSize) / configuration.Overlap;
if (width < 1)
{
return(new List <SpectralImage>());
}
float[] frames = new float[width * configuration.LogBins];
ushort[] logFrequenciesIndexes = logUtility.GenerateLogFrequenciesRanges(audioSamples.SampleRate, configuration);
float[] window = configuration.Window.GetWindow(wdftSize);
float[] samples = audioSamples.Samples;
// PIN: reverted the following FFT to use lomontFFT with managed code (not the unsafe changed made by the original author due to the issues on my computers)
// NOTE! When using Parallell.For the result becomes different from time to time
// when running in Release mode.
// Therefore make sure to use for loop instead
for (int index = 0; index < width; index++)
// Parallel.For(0, width, index =>
{
var fftArray = CopyAndWindow(samples, index * configuration.Overlap, window);
lomontFFT.RealFFT(fftArray, true);
// after the lomont realfft the fft input array will contain the FFT values
// r0, r(n/2), r1, i1, r2, i2 ...
// since the extract log bins method only uses lowBound index above 2 we can ignore the fact
// that the first and second values are "special": r0, r(n/2)
// see https://github.com/perivar/FindSimilar/blob/6b658b1c54d1504136e25e933f39b7c303da5d9e/Mirage/Fft.cs
ExtractLogBins(fftArray, logFrequenciesIndexes, configuration.LogBins, wdftSize, frames, index);
}
// );
if (configuration.Verbosity == Verbosity.Verbose)
{
var imageService = new FindSimilarImageService();
using (Image image = imageService.GetSpectrogramImage(frames, width, configuration.LogBins, width, configuration.LogBins))
{
var fileName = Path.Combine(SoundFingerprinter.DEBUG_DIRECTORY_PATH, (Path.GetFileNameWithoutExtension(audioSamples.Origin) + "_spectrogram.png"));
if (fileName != null)
{
image.Save(fileName, ImageFormat.Png);
}
}
WriteOutputUtils.WriteCSV(frames, Path.Combine(SoundFingerprinter.DEBUG_DIRECTORY_PATH, (Path.GetFileNameWithoutExtension(audioSamples.Origin) + "_frames.csv")));
}
var spectralImages = CutLogarithmizedSpectrum(frames, audioSamples.SampleRate, configuration);
if (configuration.Verbosity == Verbosity.Verbose)
{
if (spectralImages.Count > 0)
{
var spectralImageList = new List <float[]>();
foreach (var spectralImage in spectralImages)
{
spectralImageList.Add(spectralImage.Image);
}
var spectralImageArray = spectralImageList.ToArray();
WriteOutputUtils.WriteCSV(spectralImageArray, Path.Combine(SoundFingerprinter.DEBUG_DIRECTORY_PATH, (Path.GetFileNameWithoutExtension(audioSamples.Origin) + "_spectral_images.csv")), ";");
}
}
ScaleFullSpectrum(spectralImages, configuration);
return(spectralImages);
}
}
