private void ScaleFullSpectrum(IEnumerable <SpectralImage> spectralImages, SpectrogramConfig configuration)
{
Parallel.ForEach(spectralImages, image =>
{
ScaleSpectrum(image, configuration.ScalingFunction);
});
}
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 int[] GenerateLogFrequenciesRanges(int sampleRate, SpectrogramConfig configuration)
{
if (configuration.UseDynamicLogBase)
{
return(GenerateLogFrequenciesDynamicBase(sampleRate, configuration));
}
return(GenerateStaticLogFrequencies(sampleRate, configuration));
}
internal FindSimilarSpectrumService(SpectrogramConfig configuration, ILogUtility logUtility)
{
this.logUtility = logUtility;
this.lomontFFT = new Lomont.LomontFFT();
lomontFFT.A = 1;
lomontFFT.B = 1;
lomontFFT.Initialize(configuration.WdftSize);
}
private int[] GenerateLogFrequenciesDynamicBase(int sampleRate, SpectrogramConfig configuration)
{
double logBase = Math.Exp(Math.Log((float)configuration.FrequencyRange.Max / configuration.FrequencyRange.Min) / configuration.LogBins);
double mincoef = (float)configuration.WdftSize / sampleRate * configuration.FrequencyRange.Min;
int[] indexes = new int[configuration.LogBins + 1];
for (int j = 0; j < configuration.LogBins + 1; j++)
{
int start = (int)((Math.Pow(logBase, j) - 1.0) * mincoef);
indexes[j] = start + (int)mincoef;
}
return(indexes);
}
public void CustomSpectrumValuesInheritFromDefault()
{
SpectrogramConfig defaultConfiguration = SpectrogramConfig.Default;
SpectrogramConfig customConfiguration = new CustomSpectrogramConfig();
Assert.AreEqual(defaultConfiguration.ImageLength, customConfiguration.ImageLength);
Assert.AreEqual(defaultConfiguration.LogBase, customConfiguration.LogBase);
Assert.AreEqual(defaultConfiguration.LogBins, customConfiguration.LogBins);
Assert.AreEqual(defaultConfiguration.FrequencyRange.Max, customConfiguration.FrequencyRange.Max);
Assert.AreEqual(defaultConfiguration.FrequencyRange.Min, customConfiguration.FrequencyRange.Min);
Assert.AreEqual(defaultConfiguration.UseDynamicLogBase, customConfiguration.UseDynamicLogBase);
Assert.AreEqual(defaultConfiguration.WdftSize, customConfiguration.WdftSize);
Assert.AreEqual(defaultConfiguration.Overlap, customConfiguration.Overlap);
}
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));
}
private int[] GenerateStaticLogFrequencies(int sampleRate, SpectrogramConfig configuration)
{
double logMin = Math.Log(configuration.FrequencyRange.Min, configuration.LogBase);
double logMax = Math.Log(configuration.FrequencyRange.Max, configuration.LogBase);
double delta = (logMax - logMin) / configuration.LogBins;
int[] indexes = new int[configuration.LogBins + 1];
double accDelta = 0;
for (int i = 0; i <= configuration.LogBins; ++i)
{
float freq = (float)Math.Pow(configuration.LogBase, logMin + accDelta);
accDelta += delta;
indexes[i] = FrequencyToSpectrumIndex(freq, sampleRate, configuration.WdftSize); // Find the start index in array from which to start the summation
}
return(indexes);
}
protected List <SpectralImage> CutLogarithmizedSpectrum(float[][] logarithmizedSpectrum, int sampleRate, SpectrogramConfig configuration)
{
var strideBetweenConsecutiveImages = configuration.Stride;
int overlap = configuration.Overlap;
int index = (int)((float)strideBetweenConsecutiveImages.FirstStride / overlap);
int numberOfLogBins = logarithmizedSpectrum[0].Length;
var spectralImages = new List <SpectralImage>();
int width = logarithmizedSpectrum.GetLength(0);
int fingerprintImageLength = configuration.ImageLength;
int sequenceNumber = 0;
while (index + fingerprintImageLength <= width)
{
float[][] spectralImage = AllocateMemoryForFingerprintImage(fingerprintImageLength, numberOfLogBins);
for (int i = 0; i < fingerprintImageLength; i++)
{
Array.Copy(logarithmizedSpectrum[index + i], spectralImage[i], numberOfLogBins);
}
spectralImages.Add(new SpectralImage
{
Image = spectralImage,
Timestamp = index * ((double)overlap / sampleRate),
SequenceNumber = ++sequenceNumber
});
index += fingerprintImageLength + (int)((float)strideBetweenConsecutiveImages.GetNextStride() / overlap);
}
return(spectralImages);
}
public new List <SpectralImage> CutLogarithmizedSpectrum(float[][] logarithmizedSpectrum, int sampleRate, SpectrogramConfig configuration)
{
return(base.CutLogarithmizedSpectrum(logarithmizedSpectrum, sampleRate, configuration));
}
public List <SpectralImage> CutLogarithmizedSpectrum(float[] logarithmizedSpectrum, int sampleRate, SpectrogramConfig configuration)
{
var strideBetweenConsecutiveImages = configuration.Stride;
int overlap = configuration.Overlap;
int index = GetFrequencyIndexLocationOfAudioSamples(strideBetweenConsecutiveImages.FirstStride, overlap);
int numberOfLogBins = configuration.LogBins;
var spectralImages = new List <SpectralImage>();
int width = logarithmizedSpectrum.Length / numberOfLogBins;
ushort fingerprintImageLength = configuration.ImageLength;
int fullLength = configuration.ImageLength * numberOfLogBins;
uint sequenceNumber = 0;
while (index + fingerprintImageLength <= width)
{
float[] spectralImage = new float[fingerprintImageLength * numberOfLogBins];
Buffer.BlockCopy(logarithmizedSpectrum, sizeof(float) * index * numberOfLogBins, spectralImage, 0, fullLength * sizeof(float));
float startsAt = index * ((float)overlap / sampleRate);
spectralImages.Add(new SpectralImage(spectralImage, fingerprintImageLength, (ushort)numberOfLogBins, startsAt, sequenceNumber));
index += fingerprintImageLength + GetFrequencyIndexLocationOfAudioSamples(strideBetweenConsecutiveImages.NextStride, overlap);
sequenceNumber++;
}
return(spectralImages);
}
public List <SpectralImage> CutLogarithmizedSpectrum(float[][] logarithmizedSpectrum, int sampleRate, SpectrogramConfig configuration)
{
var strideBetweenConsecutiveImages = configuration.Stride;
int overlap = configuration.Overlap;
int index = GetFrequencyIndexLocationOfAudioSamples(strideBetweenConsecutiveImages.FirstStride, overlap);
int numberOfLogBins = configuration.LogBins;
var spectralImages = new List <SpectralImage>();
int width = logarithmizedSpectrum.GetLength(0);
int fingerprintImageLength = configuration.ImageLength;
int sequenceNumber = 0;
while (index + fingerprintImageLength <= width)
{
float[][] spectralImage = AllocateMemoryForFingerprintImage(fingerprintImageLength, numberOfLogBins);
for (int i = 0; i < fingerprintImageLength; i++)
{
Buffer.BlockCopy(logarithmizedSpectrum[index + i], 0, spectralImage[i], 0, numberOfLogBins * sizeof(float));
}
var startsAt = index * ((double)overlap / sampleRate);
spectralImages.Add(new SpectralImage(spectralImage, startsAt, sequenceNumber));
index += fingerprintImageLength + GetFrequencyIndexLocationOfAudioSamples(strideBetweenConsecutiveImages.NextStride, overlap);
sequenceNumber++;
}
return(spectralImages);
}
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);
}
}
