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 int[] GenerateLogFrequenciesRanges(int sampleRate, SpectrogramConfig configuration)
{
if (configuration.UseDynamicLogBase)
{
return GenerateLogFrequenciesDynamicBase(sampleRate, configuration);
}
return GenerateStaticLogFrequencies(sampleRate, configuration);
}
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 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);
float[] window = configuration.Window.GetWindow(configuration.WdftSize);
Parallel.For(0, width, i => {
float[] complexSignal = fftService.FFTForward(audioSamples.Samples, i * configuration.Overlap, configuration.WdftSize, window);
frames[i] = ExtractLogBins(complexSignal, logFrequenciesIndexes, configuration.LogBins, configuration.WdftSize);
});
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;
}
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.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;
}
