public static Tuple <BaseSonogram, AcousticEvent, double[, ], double[], double[, ]> Execute_Extraction(
AudioRecording recording,
double eventStart,
double eventEnd,
int minHz,
int maxHz,
double frameOverlap,
double backgroundThreshold,
TimeSpan segmentStartOffset)
{
//ii: MAKE SONOGRAM
SonogramConfig sonoConfig = new SonogramConfig(); //default values config
sonoConfig.SourceFName = recording.BaseName;
//sonoConfig.WindowSize = windowSize;
sonoConfig.WindowOverlap = frameOverlap;
BaseSonogram sonogram = new SpectrogramStandard(sonoConfig, recording.WavReader);
Log.WriteLine("Frames: Size={0}, Count={1}, Duration={2:f1}ms, Overlap={5:f2}%, Offset={3:f1}ms, Frames/s={4:f1}",
sonogram.Configuration.WindowSize, sonogram.FrameCount, (sonogram.FrameDuration * 1000),
(sonogram.FrameStep * 1000), sonogram.FramesPerSecond, frameOverlap);
int binCount = (int)(maxHz / sonogram.FBinWidth) - (int)(minHz / sonogram.FBinWidth) + 1;
Log.WriteIfVerbose("Freq band: {0} Hz - {1} Hz. (Freq bin count = {2})", minHz, maxHz, binCount);
//calculate the modal noise profile
double SD_COUNT = 0.0; // number of noise standard deviations used to calculate noise threshold
NoiseProfile profile = NoiseProfile.CalculateModalNoiseProfile(sonogram.Data, SD_COUNT); //calculate modal noise profile
double[] modalNoise = DataTools.filterMovingAverage(profile.NoiseMode, 7); //smooth the noise profile
//extract modal noise values of the required event
double[] noiseSubband = SpectrogramTools.ExtractModalNoiseSubband(modalNoise, minHz, maxHz, false, sonogram.NyquistFrequency, sonogram.FBinWidth);
//extract data values of the required event
double[,] target = SpectrogramTools.ExtractEvent(sonogram.Data, eventStart, eventEnd, sonogram.FrameStep,
minHz, maxHz, false, sonogram.NyquistFrequency, sonogram.FBinWidth);
// create acoustic event with defined boundaries
AcousticEvent ae = new AcousticEvent(segmentStartOffset, eventStart, eventEnd - eventStart, minHz, maxHz);
ae.SetTimeAndFreqScales(sonogram.FramesPerSecond, sonogram.FBinWidth);
//truncate noise
sonogram.Data = SNR.TruncateBgNoiseFromSpectrogram(sonogram.Data, modalNoise);
sonogram.Data = SNR.RemoveNeighbourhoodBackgroundNoise(sonogram.Data, backgroundThreshold);
double[,] targetMinusNoise = SpectrogramTools.ExtractEvent(sonogram.Data, eventStart, eventEnd, sonogram.FrameStep,
minHz, maxHz, false, sonogram.NyquistFrequency, sonogram.FBinWidth);
return(Tuple.Create(sonogram, ae, target, noiseSubband, targetMinusNoise));
}
public static double[,] GetDecibelSpectrogramNoiseReduced(AudioRecording recording, int frameSize)
{
int frameStep = frameSize;
// get decibel spectrogram
var results = DSP_Frames.ExtractEnvelopeAndAmplSpectrogram(recording.WavReader.Samples, recording.SampleRate, recording.Epsilon, frameSize, frameStep);
var spectrogram = MFCCStuff.DecibelSpectra(results.AmplitudeSpectrogram, results.WindowPower, recording.SampleRate, recording.Epsilon);
// remove background noise from spectrogram
double[] spectralDecibelBgn = NoiseProfile.CalculateBackgroundNoise(spectrogram);
spectrogram = SNR.TruncateBgNoiseFromSpectrogram(spectrogram, spectralDecibelBgn);
spectrogram = SNR.RemoveNeighbourhoodBackgroundNoise(spectrogram, nhThreshold: 3.0);
return(spectrogram);
}
/// <summary>
/// This method produces four spectrograms using four different values of neighbour hood decibel threshold.
/// It can be used for test purposes.
/// </summary>
/// <param name="deciBelSpectrogram">the noisy decibel spectrogram</param>
/// <param name="xAxisInterval">x-axis tic interval</param>
/// <param name="stepDuration">the x-axis times scale</param>
/// <param name="nyquist">max freq value</param>
/// <param name="hzInterval">y-axis frequency scale</param>
/// <returns>Image containing four sepctrograms</returns>
public static Image ModalNoiseRemovalAndGetSonograms(
double[,] deciBelSpectrogram,
TimeSpan xAxisInterval,
TimeSpan stepDuration,
int nyquist,
int hzInterval)
{
// The number of SDs above the mean for noise removal.
// Set sdCount = -0.5 becuase when sdCount >= zero, noies removal is a bit severe for environmental recordings.
var sdCount = -0.5;
var nrt = NoiseReductionType.Modal;
var tuple = SNR.NoiseReduce(deciBelSpectrogram, nrt, sdCount);
var noiseReducedSpectrogram1 = tuple.Item1;
var title = "title1";
var image1 = DrawSonogram(noiseReducedSpectrogram1, xAxisInterval, stepDuration, nyquist, hzInterval, title);
double dBThreshold = 0.0; // SPECTRAL dB THRESHOLD for smoothing background
double[,] noiseReducedSpectrogram2 = SNR.RemoveNeighbourhoodBackgroundNoise(noiseReducedSpectrogram1, dBThreshold);
title = "title2";
var image2 = DrawSonogram(noiseReducedSpectrogram2, xAxisInterval, stepDuration, nyquist, hzInterval, title);
// SPECTRAL dB THRESHOLD for smoothing background
dBThreshold = 3.0;
noiseReducedSpectrogram2 = SNR.RemoveNeighbourhoodBackgroundNoise(noiseReducedSpectrogram1, dBThreshold);
title = "title3";
var image3 = DrawSonogram(noiseReducedSpectrogram2, xAxisInterval, stepDuration, nyquist, hzInterval, title);
// SPECTRAL dB THRESHOLD for smoothing background
dBThreshold = 10.0;
noiseReducedSpectrogram2 = SNR.RemoveNeighbourhoodBackgroundNoise(noiseReducedSpectrogram1, dBThreshold);
title = "title4";
var image4 = DrawSonogram(noiseReducedSpectrogram2, xAxisInterval, stepDuration, nyquist, hzInterval, title);
var array = new Image[4];
array[0] = image1;
array[1] = image2;
array[2] = image3;
array[3] = image4;
var combinedImage = ImageTools.CombineImagesVertically(array);
return(combinedImage);
}
} // LocalPeaks()
/// <summary>
/// CALCULATEs SPECTRAL PEAK TRACKS: spectralIndices.SPT, RHZ, RVT, RPS, RNG
/// This method is only called from IndexCalulate.analysis() when the IndexCalculation Duration is less than 10 seconds,
/// because need to recalculate background noise etc.
/// Otherwise the constructor of this class is called: sptInfo = new SpectralPeakTracks(decibelSpectrogram, peakThreshold);
/// NOTE: We require a noise reduced decibel spectrogram
/// FreqBinWidth can be accessed, if required, through dspOutput1.FreqBinWidth.
/// </summary>
public static SpectralPeakTracks CalculateSpectralPeakTracks(AudioRecording recording, int sampleStart, int sampleEnd, int frameSize, bool octaveScale, double peakThreshold)
{
double epsilon = recording.Epsilon;
int sampleRate = recording.WavReader.SampleRate;
int bufferFrameCount = 2; // 2 because must allow for edge effects when using 5x5 grid to find ridges.
int ridgeBuffer = frameSize * bufferFrameCount;
var ridgeRecording = AudioRecording.GetRecordingSubsegment(recording, sampleStart, sampleEnd, ridgeBuffer);
int frameStep = frameSize;
var dspOutput = DSP_Frames.ExtractEnvelopeAndFfts(ridgeRecording, frameSize, frameStep);
// Generate the ridge SUBSEGMENT deciBel spectrogram from the SUBSEGMENT amplitude spectrogram
// i: generate the SUBSEGMENT deciBel spectrogram from the SUBSEGMENT amplitude spectrogram
double[,] decibelSpectrogram;
if (octaveScale)
{
var freqScale = new FrequencyScale(FreqScaleType.Linear125Octaves7Tones28Nyquist32000);
decibelSpectrogram = OctaveFreqScale.DecibelSpectra(dspOutput.AmplitudeSpectrogram, dspOutput.WindowPower, sampleRate, epsilon, freqScale);
}
else
{
decibelSpectrogram = MFCCStuff.DecibelSpectra(dspOutput.AmplitudeSpectrogram, dspOutput.WindowPower, sampleRate, epsilon);
}
// calculate the noise profile
var spectralDecibelBgn = NoiseProfile.CalculateBackgroundNoise(decibelSpectrogram);
decibelSpectrogram = SNR.TruncateBgNoiseFromSpectrogram(decibelSpectrogram, spectralDecibelBgn);
double nhDecibelThreshold = 2.0; // SPECTRAL dB THRESHOLD for smoothing background
decibelSpectrogram = SNR.RemoveNeighbourhoodBackgroundNoise(decibelSpectrogram, nhDecibelThreshold);
// thresholds in decibels
// double frameStepDuration = frameStep / (double)sampleRate; // fraction of a second
// TimeSpan frameStepTimeSpan = TimeSpan.FromTicks((long)(frameStepDuration * TimeSpan.TicksPerSecond));
var sptInfo = new SpectralPeakTracks(decibelSpectrogram, peakThreshold);
return(sptInfo);
}
//////public static IndexCalculateResult Analysis(
public static SpectralIndexValuesForContentDescription Analysis(
AudioRecording recording,
TimeSpan segmentOffsetTimeSpan,
int sampleRateOfOriginalAudioFile,
bool returnSonogramInfo = false)
{
// returnSonogramInfo = true; // if debugging
double epsilon = recording.Epsilon;
int sampleRate = recording.WavReader.SampleRate;
//var segmentDuration = TimeSpan.FromSeconds(recording.WavReader.Time.TotalSeconds);
var indexCalculationDuration = TimeSpan.FromSeconds(ContentSignatures.IndexCalculationDurationInSeconds);
// Get FRAME parameters for the calculation of Acoustic Indices
int frameSize = ContentSignatures.FrameSize;
int frameStep = frameSize; // that is, windowOverlap = zero
double frameStepDuration = frameStep / (double)sampleRate; // fraction of a second
var frameStepTimeSpan = TimeSpan.FromTicks((long)(frameStepDuration * TimeSpan.TicksPerSecond));
// INITIALISE a RESULTS STRUCTURE TO return
// initialize a result object in which to store SummaryIndexValues and SpectralIndexValues etc.
var config = new IndexCalculateConfig(); // sets some default values
int freqBinCount = frameSize / 2;
var indexProperties = GetIndexProperties();
////////var result = new IndexCalculateResult(freqBinCount, indexProperties, indexCalculationDuration, segmentOffsetTimeSpan, config);
var spectralIndices = new SpectralIndexValuesForContentDescription();
///////result.SummaryIndexValues = null;
///////SpectralIndexValues spectralIndices = result.SpectralIndexValues;
// set up default spectrogram to return
///////result.Sg = returnSonogramInfo ? GetSonogram(recording, windowSize: 1024) : null;
///////result.Hits = null;
///////result.TrackScores = new List<Plot>();
// ################################## FINISHED SET-UP
// ################################## NOW GET THE AMPLITUDE SPECTROGRAM
// EXTRACT ENVELOPE and SPECTROGRAM FROM RECORDING SEGMENT
// Note that the amplitude spectrogram has had the DC bin removed. i.e. has only 256 columns.
var dspOutput1 = DSP_Frames.ExtractEnvelopeAndFfts(recording, frameSize, frameStep);
var amplitudeSpectrogram = dspOutput1.AmplitudeSpectrogram;
// (B) ################################## EXTRACT OSC SPECTRAL INDEX DIRECTLY FROM THE RECORDING ##################################
// Get the oscillation spectral index OSC separately from signal because need a different frame size etc.
var sampleLength = Oscillations2014.DefaultSampleLength;
var frameLength = Oscillations2014.DefaultFrameLength;
var sensitivity = Oscillations2014.DefaultSensitivityThreshold;
var spectralIndexShort = Oscillations2014.GetSpectralIndex_Osc(recording, frameLength, sampleLength, sensitivity);
// double length of the vector because want to work with 256 element vector for spectrogram purposes
spectralIndices.OSC = DataTools.VectorDoubleLengthByAverageInterpolation(spectralIndexShort);
// (C) ################################## EXTRACT SPECTRAL INDICES FROM THE AMPLITUDE SPECTROGRAM ##################################
// IFF there has been UP-SAMPLING, calculate bin of the original audio nyquist. this will be less than SR/2.
// original sample rate can be anything 11.0-44.1 kHz.
int originalNyquist = sampleRateOfOriginalAudioFile / 2;
// if up-sampling has been done
if (dspOutput1.NyquistFreq > originalNyquist)
{
dspOutput1.NyquistFreq = originalNyquist;
dspOutput1.NyquistBin = (int)Math.Floor(originalNyquist / dspOutput1.FreqBinWidth); // note that bin width does not change
}
// ii: CALCULATE THE ACOUSTIC COMPLEXITY INDEX
spectralIndices.ACI = AcousticComplexityIndex.CalculateAci(amplitudeSpectrogram);
// iii: CALCULATE the H(t) or Temporal ENTROPY Spectrum and then reverse the values i.e. calculate 1-Ht for energy concentration
double[] temporalEntropySpectrum = AcousticEntropy.CalculateTemporalEntropySpectrum(amplitudeSpectrogram);
for (int i = 0; i < temporalEntropySpectrum.Length; i++)
{
temporalEntropySpectrum[i] = 1 - temporalEntropySpectrum[i];
}
spectralIndices.ENT = temporalEntropySpectrum;
// (C) ################################## EXTRACT SPECTRAL INDICES FROM THE DECIBEL SPECTROGRAM ##################################
// i: Convert amplitude spectrogram to decibels and calculate the dB background noise profile
double[,] decibelSpectrogram = MFCCStuff.DecibelSpectra(dspOutput1.AmplitudeSpectrogram, dspOutput1.WindowPower, sampleRate, epsilon);
double[] spectralDecibelBgn = NoiseProfile.CalculateBackgroundNoise(decibelSpectrogram);
spectralIndices.BGN = spectralDecibelBgn;
// ii: Calculate the noise reduced decibel spectrogram derived from segment recording.
// REUSE the var decibelSpectrogram but this time using dspOutput1.
decibelSpectrogram = MFCCStuff.DecibelSpectra(dspOutput1.AmplitudeSpectrogram, dspOutput1.WindowPower, sampleRate, epsilon);
decibelSpectrogram = SNR.TruncateBgNoiseFromSpectrogram(decibelSpectrogram, spectralDecibelBgn);
decibelSpectrogram = SNR.RemoveNeighbourhoodBackgroundNoise(decibelSpectrogram, nhThreshold: 2.0);
// iii: CALCULATE noise reduced AVERAGE DECIBEL SPECTRUM
spectralIndices.PMN = SpectrogramTools.CalculateAvgDecibelSpectrumFromDecibelSpectrogram(decibelSpectrogram);
// ######################################################################################################################################################
// iv: CALCULATE SPECTRAL COVER. NOTE: at this point, decibelSpectrogram is noise reduced. All values >= 0.0
// FreqBinWidth can be accessed, if required, through dspOutput1.FreqBinWidth
// dB THRESHOLD for calculating spectral coverage
double dBThreshold = ActivityAndCover.DefaultActivityThresholdDb;
// Calculate lower and upper boundary bin ids.
// Boundary between low & mid frequency bands is to avoid low freq bins containing anthropogenic noise. These biased index values away from bio-phony.
int midFreqBound = config.MidFreqBound;
int lowFreqBound = config.LowFreqBound;
int lowerBinBound = (int)Math.Ceiling(lowFreqBound / dspOutput1.FreqBinWidth);
int middleBinBound = (int)Math.Ceiling(midFreqBound / dspOutput1.FreqBinWidth);
var spActivity = ActivityAndCover.CalculateSpectralEvents(decibelSpectrogram, dBThreshold, frameStepTimeSpan, lowerBinBound, middleBinBound);
//spectralIndices.CVR = spActivity.CoverSpectrum;
spectralIndices.EVN = spActivity.EventSpectrum;
///////result.TrackScores = null;
///////return result;
return(spectralIndices);
} // end calculation of Six Spectral Indices
/// <summary>
/// Calculate summary statistics for supplied temporal and spectral targets.
/// </summary>
/// <remarks>
/// The acoustic statistics calculated in this method are based on methods outlined in
/// "Acoustic classification of multiple simultaneous bird species: A multi-instance multi-label approach",
/// by Forrest Briggs, Balaji Lakshminarayanan, Lawrence Neal, Xiaoli Z.Fern, Raviv Raich, Sarah J.K.Hadley, Adam S. Hadley, Matthew G. Betts, et al.
/// The Journal of the Acoustical Society of America v131, pp4640 (2012); doi: http://dx.doi.org/10.1121/1.4707424
/// ..
/// The Briggs feature are calculated from the column (freq bin) and row (frame) sums of the extracted spectrogram.
/// 1. Gini Index for frame and bin sums. A measure of dispersion. Problem with gini is that its value is dependent on the row or column count.
/// We use entropy instead because value not dependent on row or column count because it is normalized.
/// For the following meausres of k-central moments, the freq and time values are normalized in 0,1 to width of the event.
/// 2. freq-mean
/// 3. freq-variance
/// 4. freq-skew and kurtosis
/// 5. time-mean
/// 6. time-variance
/// 7. time-skew and kurtosis
/// 8. freq-max (normalized)
/// 9. time-max (normalized)
/// 10. Briggs et al also calculate a 16 value histogram of gradients for each event mask. We do not do that here although we could.
/// ...
/// NOTE 1: There are differences between our method of noise reduction and Briggs. Briggs does not convert to decibels
/// and instead works with power values. He obtains a noise profile from the 20% of frames having the lowest energy sum.
/// NOTE 2: To NormaliseMatrixValues for noise, they divide the actual energy by the noise value. This is equivalent to subtraction when working in decibels.
/// There are advantages and disadvantages to Briggs method versus ours. In our case, we hve to convert decibel values back to
/// energy values when calculating the statistics for the extracted acoustic event.
/// NOTE 3: We do not calculate the higher central moments of the time/frequency profiles, i.e. skew and kurtosis.
/// Ony mean and standard deviation.
/// ..
/// NOTE 4: This method assumes that the passed event occurs totally within the passed recording,
/// AND that the passed recording is of sufficient duration to obtain reliable BGN noise profile
/// BUT not so long as to cause memory constipation.
/// </remarks>
/// <param name="recording">as type AudioRecording which contains the event</param>
/// <param name="temporalTarget">Both start and end bounds - relative to the supplied recording</param>
/// <param name="spectralTarget">both bottom and top bounds in Hertz</param>
/// <param name="config">parameters that determine the outcome of the analysis</param>
/// <param name="segmentStartOffset">How long since the start of the recording this event occurred</param>
/// <returns>an instance of EventStatistics</returns>
public static EventStatistics AnalyzeAudioEvent(
AudioRecording recording,
Range <TimeSpan> temporalTarget,
Range <double> spectralTarget,
EventStatisticsConfiguration config,
TimeSpan segmentStartOffset)
{
var stats = new EventStatistics
{
EventStartSeconds = temporalTarget.Minimum.TotalSeconds,
EventEndSeconds = temporalTarget.Maximum.TotalSeconds,
LowFrequencyHertz = spectralTarget.Minimum,
HighFrequencyHertz = spectralTarget.Maximum,
SegmentDurationSeconds = recording.Duration.TotalSeconds,
SegmentStartSeconds = segmentStartOffset.TotalSeconds,
};
// temporal target is supplied relative to recording, but not the supplied audio segment
// shift coordinates relative to segment
var localTemporalTarget = temporalTarget.Shift(-segmentStartOffset);
if (!recording
.Duration
.AsRangeFromZero(Topology.Inclusive)
.Contains(localTemporalTarget))
{
stats.Error = true;
stats.ErrorMessage =
$"Audio not long enough ({recording.Duration}) to analyze target ({localTemporalTarget})";
return(stats);
}
// convert recording to spectrogram
int sampleRate = recording.SampleRate;
double epsilon = recording.Epsilon;
// extract the spectrogram
var dspOutput1 = DSP_Frames.ExtractEnvelopeAndFfts(recording, config.FrameSize, config.FrameStep);
double hertzBinWidth = dspOutput1.FreqBinWidth;
var stepDurationInSeconds = config.FrameStep / (double)sampleRate;
var startFrame = (int)Math.Ceiling(localTemporalTarget.Minimum.TotalSeconds / stepDurationInSeconds);
// subtract 1 frame because want to end before start of end point.
var endFrame = (int)Math.Floor(localTemporalTarget.Maximum.TotalSeconds / stepDurationInSeconds) - 1;
var bottomBin = (int)Math.Floor(spectralTarget.Minimum / hertzBinWidth);
var topBin = (int)Math.Ceiling(spectralTarget.Maximum / hertzBinWidth);
// Events can have their high value set to the nyquist.
// Since the submatrix call below uses an inclusive upper bound an index out of bounds exception occurs in
// these cases. So we just ask for the bin below.
if (topBin >= config.FrameSize / 2)
{
topBin = (config.FrameSize / 2) - 1;
}
// Convert amplitude spectrogram to deciBels and calculate the dB background noise profile
double[,] decibelSpectrogram = MFCCStuff.DecibelSpectra(dspOutput1.AmplitudeSpectrogram, dspOutput1.WindowPower, sampleRate, epsilon);
double[] spectralDecibelBgn = NoiseProfile.CalculateBackgroundNoise(decibelSpectrogram);
decibelSpectrogram = SNR.TruncateBgNoiseFromSpectrogram(decibelSpectrogram, spectralDecibelBgn);
decibelSpectrogram = SNR.RemoveNeighbourhoodBackgroundNoise(decibelSpectrogram, nhThreshold: 2.0);
// extract the required acoustic event
var eventMatrix = MatrixTools.Submatrix(decibelSpectrogram, startFrame, bottomBin, endFrame, topBin);
// Get the SNR of the event. This is just the max value in the matrix because noise reduced
MatrixTools.MinMax(eventMatrix, out _, out double max);
stats.SnrDecibels = max;
// Now need to convert event matrix back to energy values before calculating other statistics
eventMatrix = MatrixTools.Decibels2Power(eventMatrix);
var columnAverages = MatrixTools.GetColumnAverages(eventMatrix);
var rowAverages = MatrixTools.GetRowAverages(eventMatrix);
// calculate the mean and temporal standard deviation in decibels
NormalDist.AverageAndSD(rowAverages, out double mean, out double stddev);
stats.MeanDecibels = 10 * Math.Log10(mean);
stats.TemporalStdDevDecibels = 10 * Math.Log10(stddev);
// calculate the frequency standard deviation in decibels
NormalDist.AverageAndSD(columnAverages, out mean, out stddev);
stats.FreqBinStdDevDecibels = 10 * Math.Log10(stddev);
// calculate relative location of the temporal maximum
int maxRowId = DataTools.GetMaxIndex(rowAverages);
stats.TemporalMaxRelative = maxRowId / (double)rowAverages.Length;
// calculate the entropy dispersion/concentration indices
stats.TemporalEnergyDistribution = 1 - DataTools.EntropyNormalised(rowAverages);
stats.SpectralEnergyDistribution = 1 - DataTools.EntropyNormalised(columnAverages);
// calculate the spectral centroid and the dominant frequency
double binCentroid = CalculateSpectralCentroid(columnAverages);
stats.SpectralCentroid = (int)Math.Round(hertzBinWidth * binCentroid) + (int)spectralTarget.Minimum;
int maxColumnId = DataTools.GetMaxIndex(columnAverages);
stats.DominantFrequency = (int)Math.Round(hertzBinWidth * maxColumnId) + (int)spectralTarget.Minimum;
// remainder of this method is to produce debugging images. Can comment out when not debugging.
/*
* var normalisedIndex = DataTools.NormaliseMatrixValues(columnAverages);
* var image4 = GraphsAndCharts.DrawGraph("columnSums", normalisedIndex, 100);
* string path4 = @"C:\SensorNetworks\Output\Sonograms\UnitTestSonograms\columnSums.png";
* image4.Save(path4);
* normalisedIndex = DataTools.NormaliseMatrixValues(rowAverages);
* image4 = GraphsAndCharts.DrawGraph("rowSums", normalisedIndex, 100);
* path4 = @"C:\SensorNetworks\Output\Sonograms\UnitTestSonograms\rowSums.png";
* image4.Save(path4);
*/
return(stats);
}
/// <summary>
/// Calculates the following spectrograms as per settings in the Images array in the config file: Towsey.SpectrogramGenerator.yml:
/// Waveform.
/// DecibelSpectrogram.
/// DecibelSpectrogramNoiseReduced.
/// CepstralSpectrogram.
/// DifferenceSpectrogram.
/// AmplitudeSpectrogramLocalContrastNormalization.
/// Experimental.
/// Comment the config.yml file with a hash, those spectrograms that are not required.
/// </summary>
/// <param name="sourceRecording">The name of the original recording.</param>
/// <param name="config">Contains parameter info to make spectrograms.</param>
/// <param name="sourceRecordingName">.Name of source recording. Required only spectrogram labels.</param>
public static AudioToSonogramResult GenerateSpectrogramImages(
FileInfo sourceRecording,
SpectrogramGeneratorConfig config,
string sourceRecordingName)
{
//int signalLength = recordingSegment.WavReader.GetChannel(0).Length;
var recordingSegment = new AudioRecording(sourceRecording.FullName);
int sampleRate = recordingSegment.WavReader.SampleRate;
var result = new AudioToSonogramResult();
var requestedImageTypes = config.Images ?? new[] { SpectrogramImageType.DecibelSpectrogram };
var @do = requestedImageTypes.ToHashSet();
int frameSize = config.GetIntOrNull("FrameLength") ?? 512;
int frameStep = config.GetIntOrNull("FrameStep") ?? 441;
// must calculate this because used later on.
double frameOverlap = (frameSize - frameStep) / (double)frameSize;
// Default noiseReductionType = Standard
var bgNoiseThreshold = config.BgNoiseThreshold;
// threshold for drawing the difference spectrogram
var differenceThreshold = config.DifferenceThreshold;
// EXTRACT ENVELOPE and SPECTROGRAM FROM RECORDING SEGMENT
var dspOutput1 = DSP_Frames.ExtractEnvelopeAndFfts(recordingSegment, frameSize, frameStep);
var sonoConfig = new SonogramConfig()
{
epsilon = recordingSegment.Epsilon,
SampleRate = sampleRate,
WindowSize = frameSize,
WindowStep = frameStep,
WindowOverlap = frameOverlap,
WindowPower = dspOutput1.WindowPower,
Duration = recordingSegment.Duration,
NoiseReductionType = NoiseReductionType.Standard,
NoiseReductionParameter = bgNoiseThreshold,
};
var images = new Dictionary <SpectrogramImageType, Image <Rgb24> >(requestedImageTypes.Length);
// IMAGE 1) draw the WAVEFORM
if (@do.Contains(Waveform))
{
var minValues = dspOutput1.MinFrameValues;
var maxValues = dspOutput1.MaxFrameValues;
int height = config.WaveformHeight;
var waveformImage = GetWaveformImage(minValues, maxValues, height);
// add in the title bar and time scales.
string title =
$"WAVEFORM - {sourceRecordingName} (min value={dspOutput1.MinSignalValue:f3}, max value={dspOutput1.MaxSignalValue:f3})";
var titleBar = BaseSonogram.DrawTitleBarOfGrayScaleSpectrogram(
title,
waveformImage.Width,
ImageTags[Waveform]);
var startTime = TimeSpan.Zero;
var xAxisTicInterval = TimeSpan.FromSeconds(1);
TimeSpan xAxisPixelDuration = TimeSpan.FromSeconds(frameStep / (double)sampleRate);
var labelInterval = TimeSpan.FromSeconds(5);
waveformImage = BaseSonogram.FrameSonogram(
waveformImage,
titleBar,
startTime,
xAxisTicInterval,
xAxisPixelDuration,
labelInterval);
images.Add(Waveform, waveformImage);
}
// Draw various decibel spectrograms
var decibelTypes = new[] { SpectrogramImageType.DecibelSpectrogram, DecibelSpectrogramNoiseReduced, DifferenceSpectrogram, Experimental };
if (@do.Overlaps(decibelTypes))
{
// disable noise removal for first two spectrograms
var disabledNoiseReductionType = sonoConfig.NoiseReductionType;
sonoConfig.NoiseReductionType = NoiseReductionType.None;
//Get the decibel spectrogram
var decibelSpectrogram = new SpectrogramStandard(sonoConfig, dspOutput1.AmplitudeSpectrogram);
result.DecibelSpectrogram = decibelSpectrogram;
double[,] dbSpectrogramData = (double[, ])decibelSpectrogram.Data.Clone();
// IMAGE 2) Display the DecibelSpectrogram
if (@do.Contains(SpectrogramImageType.DecibelSpectrogram))
{
images.Add(
SpectrogramImageType.DecibelSpectrogram,
decibelSpectrogram.GetImageFullyAnnotated(
$"DECIBEL SPECTROGRAM ({sourceRecordingName})",
ImageTags[SpectrogramImageType.DecibelSpectrogram]));
}
if (@do.Overlaps(new[] { DecibelSpectrogramNoiseReduced, Experimental, CepstralSpectrogram }))
{
sonoConfig.NoiseReductionType = disabledNoiseReductionType;
sonoConfig.NoiseReductionParameter = bgNoiseThreshold;
double[] spectralDecibelBgn = NoiseProfile.CalculateBackgroundNoise(decibelSpectrogram.Data);
decibelSpectrogram.Data =
SNR.TruncateBgNoiseFromSpectrogram(decibelSpectrogram.Data, spectralDecibelBgn);
decibelSpectrogram.Data =
SNR.RemoveNeighbourhoodBackgroundNoise(decibelSpectrogram.Data, nhThreshold: bgNoiseThreshold);
// IMAGE 3) DecibelSpectrogram - noise reduced
if (@do.Contains(DecibelSpectrogramNoiseReduced))
{
images.Add(
DecibelSpectrogramNoiseReduced,
decibelSpectrogram.GetImageFullyAnnotated(
$"DECIBEL SPECTROGRAM + Lamel noise subtraction. ({sourceRecordingName})",
ImageTags[DecibelSpectrogramNoiseReduced]));
}
// IMAGE 4) EXPERIMENTAL Spectrogram
if (@do.Contains(Experimental))
{
sonoConfig.NoiseReductionType = disabledNoiseReductionType;
images.Add(
Experimental,
GetDecibelSpectrogram_Ridges(
dbSpectrogramData,
decibelSpectrogram,
sourceRecordingName));
}
}
// IMAGE 5) draw difference spectrogram. This is derived from the original decibel spectrogram
if (@do.Contains(DifferenceSpectrogram))
{
//var differenceThreshold = configInfo.GetDoubleOrNull("DifferenceThreshold") ?? 3.0;
var differenceImage = GetDifferenceSpectrogram(dbSpectrogramData, differenceThreshold);
differenceImage = BaseSonogram.GetImageAnnotatedWithLinearHertzScale(
differenceImage,
sampleRate,
frameStep,
$"DECIBEL DIFFERENCE SPECTROGRAM ({sourceRecordingName})",
ImageTags[DifferenceSpectrogram]);
images.Add(DifferenceSpectrogram, differenceImage);
}
}
// IMAGE 6) Cepstral Spectrogram
if (@do.Contains(CepstralSpectrogram))
{
images.Add(
CepstralSpectrogram,
GetCepstralSpectrogram(sonoConfig, recordingSegment, sourceRecordingName));
}
// IMAGE 7) AmplitudeSpectrogram_LocalContrastNormalization
if (@do.Contains(AmplitudeSpectrogramLocalContrastNormalization))
{
var neighborhoodSeconds = config.NeighborhoodSeconds;
var lcnContrastParameter = config.LcnContrastLevel;
images.Add(
AmplitudeSpectrogramLocalContrastNormalization,
GetLcnSpectrogram(
sonoConfig,
recordingSegment,
sourceRecordingName,
neighborhoodSeconds,
lcnContrastParameter));
}
// now pick and combine images in order user specified
var sortedImages = requestedImageTypes.Select(x => images[x]);
// COMBINE THE SPECTROGRAM IMAGES
result.CompositeImage = ImageTools.CombineImagesVertically(sortedImages.ToArray());
return(result);
}
/// <summary>
/// Calculates the following spectrograms as per content of config.yml file:
/// Waveform: true.
/// DifferenceSpectrogram: true.
/// DecibelSpectrogram: true.
/// DecibelSpectrogram_NoiseReduced: true.
/// DecibelSpectrogram_Ridges: true.
/// AmplitudeSpectrogram_LocalContrastNormalization: true.
/// SoxSpectrogram: false.
/// Experimental: true.
/// </summary>
/// <param name="sourceRecording">The name of the original recording.</param>
/// <param name="configInfo">Contains parameter info to make spectrograms.</param>
/// <param name="sourceRecordingName">.Name of source recording. Required only spectrogram labels.</param>
public static AudioToSonogramResult GenerateSpectrogramImages(
FileInfo sourceRecording,
AnalyzerConfig configInfo,
string sourceRecordingName)
{
//int signalLength = recordingSegment.WavReader.GetChannel(0).Length;
var recordingSegment = new AudioRecording(sourceRecording.FullName);
int sampleRate = recordingSegment.WavReader.SampleRate;
var result = new AudioToSonogramResult();
// init the image stack
var list = new List <Image>();
bool doWaveForm = configInfo.GetBoolOrNull("Waveform") ?? false;
bool doDecibelSpectrogram = configInfo.GetBoolOrNull("DecibelSpectrogram") ?? false;
bool doNoiseReducedSpectrogram = configInfo.GetBoolOrNull("DecibelSpectrogram_NoiseReduced") ?? true;
bool doDifferenceSpectrogram = configInfo.GetBoolOrNull("DifferenceSpectrogram") ?? false;
bool doLcnSpectrogram = configInfo.GetBoolOrNull("AmplitudeSpectrogram_LocalContrastNormalization") ?? false;
bool doCepstralSpectrogram = configInfo.GetBoolOrNull("CepstralSpectrogram") ?? false;
bool doExperimentalSpectrogram = configInfo.GetBoolOrNull("Experimental") ?? false;
//Don't do SOX spectrogram.
//bool doSoxSpectrogram = configInfo.GetBool("SoxSpectrogram");
int frameSize = configInfo.GetIntOrNull("FrameLength") ?? 512;
int frameStep = configInfo.GetIntOrNull("FrameStep") ?? 0;
// must calculate this because used later on.
double frameOverlap = (frameSize - frameStep) / (double)frameSize;
// Default noiseReductionType = Standard
var bgNoiseThreshold = configInfo.GetDoubleOrNull("BgNoiseThreshold") ?? 3.0;
// EXTRACT ENVELOPE and SPECTROGRAM FROM RECORDING SEGMENT
var dspOutput1 = DSP_Frames.ExtractEnvelopeAndFfts(recordingSegment, frameSize, frameStep);
var sonoConfig = new SonogramConfig()
{
epsilon = recordingSegment.Epsilon,
SampleRate = sampleRate,
WindowSize = frameSize,
WindowStep = frameStep,
WindowOverlap = frameOverlap,
WindowPower = dspOutput1.WindowPower,
Duration = recordingSegment.Duration,
NoiseReductionType = NoiseReductionType.Standard,
NoiseReductionParameter = bgNoiseThreshold,
};
// IMAGE 1) draw the WAVEFORM
if (doWaveForm)
{
var minValues = dspOutput1.MinFrameValues;
var maxValues = dspOutput1.MaxFrameValues;
int height = configInfo.GetIntOrNull("WaveformHeight") ?? 180;
var waveformImage = GetWaveformImage(minValues, maxValues, height);
// add in the title bar and time scales.
string title = $"WAVEFORM - {sourceRecordingName} (min value={dspOutput1.MinSignalValue:f3}, max value={dspOutput1.MaxSignalValue:f3})";
var titleBar = BaseSonogram.DrawTitleBarOfGrayScaleSpectrogram(title, waveformImage.Width);
var startTime = TimeSpan.Zero;
var xAxisTicInterval = TimeSpan.FromSeconds(1);
TimeSpan xAxisPixelDuration = TimeSpan.FromSeconds(frameStep / (double)sampleRate);
var labelInterval = TimeSpan.FromSeconds(5);
waveformImage = BaseSonogram.FrameSonogram(waveformImage, titleBar, startTime, xAxisTicInterval, xAxisPixelDuration, labelInterval);
list.Add(waveformImage);
}
// Draw various decibel spectrograms
if (doDecibelSpectrogram || doNoiseReducedSpectrogram || doDifferenceSpectrogram || doExperimentalSpectrogram)
{
// disable noise removal for first spectrogram
var disabledNoiseReductionType = sonoConfig.NoiseReductionType;
sonoConfig.NoiseReductionType = NoiseReductionType.None;
//Get the decibel spectrogram
var decibelSpectrogram = new SpectrogramStandard(sonoConfig, dspOutput1.AmplitudeSpectrogram);
result.DecibelSpectrogram = decibelSpectrogram;
double[,] dbSpectrogramData = (double[, ])decibelSpectrogram.Data.Clone();
// IMAGE 2) DecibelSpectrogram
if (doDecibelSpectrogram)
{
var image3 = decibelSpectrogram.GetImageFullyAnnotated($"DECIBEL SPECTROGRAM ({sourceRecordingName})");
list.Add(image3);
}
if (doNoiseReducedSpectrogram || doExperimentalSpectrogram || doDifferenceSpectrogram)
{
sonoConfig.NoiseReductionType = disabledNoiseReductionType;
sonoConfig.NoiseReductionParameter = bgNoiseThreshold;
double[] spectralDecibelBgn = NoiseProfile.CalculateBackgroundNoise(decibelSpectrogram.Data);
decibelSpectrogram.Data = SNR.TruncateBgNoiseFromSpectrogram(decibelSpectrogram.Data, spectralDecibelBgn);
decibelSpectrogram.Data = SNR.RemoveNeighbourhoodBackgroundNoise(decibelSpectrogram.Data, nhThreshold: bgNoiseThreshold);
// IMAGE 3) DecibelSpectrogram - noise reduced
if (doNoiseReducedSpectrogram)
{
var image4 = decibelSpectrogram.GetImageFullyAnnotated($"DECIBEL SPECTROGRAM + Lamel noise subtraction. ({sourceRecordingName})");
list.Add(image4);
}
// IMAGE 4) EXPERIMENTAL Spectrogram
if (doExperimentalSpectrogram)
{
sonoConfig.NoiseReductionType = disabledNoiseReductionType;
var image5 = GetDecibelSpectrogram_Ridges(dbSpectrogramData, decibelSpectrogram, sourceRecordingName);
list.Add(image5);
}
// IMAGE 5) draw difference spectrogram
if (doDifferenceSpectrogram)
{
var differenceThreshold = configInfo.GetDoubleOrNull("DifferenceThreshold") ?? 3.0;
var image6 = GetDifferenceSpectrogram(dbSpectrogramData, differenceThreshold);
image6 = BaseSonogram.GetImageAnnotatedWithLinearHertzScale(image6, sampleRate, frameStep, $"DECIBEL DIFFERENCE SPECTROGRAM ({sourceRecordingName})");
list.Add(image6);
}
}
}
// IMAGE 6) Cepstral Spectrogram
if (doCepstralSpectrogram)
{
var image6 = GetCepstralSpectrogram(sonoConfig, recordingSegment, sourceRecordingName);
list.Add(image6);
}
// 7) AmplitudeSpectrogram_LocalContrastNormalization
if (doLcnSpectrogram)
{
var neighbourhoodSeconds = configInfo.GetDoubleOrNull("NeighbourhoodSeconds") ?? 0.5;
var lcnContrastParameter = configInfo.GetDoubleOrNull("LcnContrastLevel") ?? 0.4;
var image8 = GetLcnSpectrogram(sonoConfig, recordingSegment, sourceRecordingName, neighbourhoodSeconds, lcnContrastParameter);
list.Add(image8);
}
// 8) SOX SPECTROGRAM
//if (doSoxSpectrogram)
//{
//Log.Warn("SoX spectrogram set to true but is ignored when running as an IAnalyzer");
// The following parameters were once used to implement a sox spectrogram.
//bool makeSoxSonogram = configuration.GetBoolOrNull(AnalysisKeys.MakeSoxSonogram) ?? false;
//configDict[AnalysisKeys.SonogramTitle] = configuration[AnalysisKeys.SonogramTitle] ?? "Sonogram";
//configDict[AnalysisKeys.SonogramComment] = configuration[AnalysisKeys.SonogramComment] ?? "Sonogram produced using SOX";
//configDict[AnalysisKeys.SonogramColored] = configuration[AnalysisKeys.SonogramColored] ?? "false";
//configDict[AnalysisKeys.SonogramQuantisation] = configuration[AnalysisKeys.SonogramQuantisation] ?? "128";
//configDict[AnalysisKeys.AddTimeScale] = configuration[AnalysisKeys.AddTimeScale] ?? "true";
//configDict[AnalysisKeys.AddAxes] = configuration[AnalysisKeys.AddAxes] ?? "true";
//configDict[AnalysisKeys.AddSegmentationTrack] = configuration[AnalysisKeys.AddSegmentationTrack] ?? "true";
// var soxFile = new FileInfo(Path.Combine(output.FullName, sourceName + "SOX.png"));
// SpectrogramTools.MakeSonogramWithSox(sourceRecording, configDict, path2SoxSpectrogram);
// list.Add(image7);
//}
// COMBINE THE SPECTROGRAM IMAGES
result.CompositeImage = ImageTools.CombineImagesVertically(list);
return(result);
}
public static IndexCalculateResult Analysis(
AudioRecording recording,
TimeSpan subsegmentOffsetTimeSpan,
Dictionary <string, IndexProperties> indexProperties,
int sampleRateOfOriginalAudioFile,
TimeSpan segmentStartOffset,
IndexCalculateConfig config,
bool returnSonogramInfo = false)
{
// returnSonogramInfo = true; // if debugging
double epsilon = recording.Epsilon;
int signalLength = recording.WavReader.GetChannel(0).Length;
int sampleRate = recording.WavReader.SampleRate;
var segmentDuration = TimeSpan.FromSeconds(recording.WavReader.Time.TotalSeconds);
var indexCalculationDuration = config.IndexCalculationDurationTimeSpan;
int nyquist = sampleRate / 2;
// Get FRAME parameters for the calculation of Acoustic Indices
//WARNING: DO NOT USE Frame Overlap when calculating acoustic indices.
// It yields ACI, BGN, POW and EVN results that are significantly different from the default.
// I have not had time to check if the difference is meaningful. Best to avoid.
//int frameSize = (int?)config[AnalysisKeys.FrameLength] ?? IndexCalculateConfig.DefaultWindowSize;
int frameSize = config.FrameLength;
int frameStep = frameSize; // that is, windowOverlap = zero
double frameStepDuration = frameStep / (double)sampleRate; // fraction of a second
var frameStepTimeSpan = TimeSpan.FromTicks((long)(frameStepDuration * TimeSpan.TicksPerSecond));
int midFreqBound = config.MidFreqBound;
int lowFreqBound = config.LowFreqBound;
int freqBinCount = frameSize / 2;
// double freqBinWidth = recording.Nyquist / (double)freqBinCount;
// get duration in seconds and sample count and frame count
double subsegmentDurationInSeconds = indexCalculationDuration.TotalSeconds;
int subsegmentSampleCount = (int)(subsegmentDurationInSeconds * sampleRate);
double subsegmentFrameCount = subsegmentSampleCount / (double)frameStep;
subsegmentFrameCount = (int)Math.Ceiling(subsegmentFrameCount);
// In order not to lose the last fractional frame, round up the frame number
// and get the exact number of samples in the integer number of frames.
// Do this because when IndexCalculationDuration = 100ms, the number of frames is only 8.
subsegmentSampleCount = (int)(subsegmentFrameCount * frameStep);
// get start and end samples of the subsegment and noise segment
double localOffsetInSeconds = subsegmentOffsetTimeSpan.TotalSeconds - segmentStartOffset.TotalSeconds;
int startSample = (int)(localOffsetInSeconds * sampleRate);
int endSample = startSample + subsegmentSampleCount - 1;
// Default behaviour: set SUBSEGMENT = total recording
var subsegmentRecording = recording;
// But if the indexCalculationDuration < segmentDuration
if (indexCalculationDuration < segmentDuration)
{
// minimum samples needed to calculate acoustic indices. This value was chosen somewhat arbitrarily.
// It allowes for case where IndexCalculationDuration = 100ms which is approx 8 frames
int minimumViableSampleCount = frameSize * 8;
int availableSignal = signalLength - startSample;
// if (the required audio is beyond recording OR insufficient for analysis) then backtrack.
if (availableSignal < minimumViableSampleCount)
{
// Back-track so we can fill a whole result.
// This is a silent correction, equivalent to having a segment overlap for the last segment.
var oldStart = startSample;
startSample = signalLength - subsegmentSampleCount;
endSample = signalLength;
Logger.Trace(" Backtrack subsegment to fill missing data from imperfect audio cuts because not enough samples available. " + (oldStart - startSample) + " samples overlap.");
}
var subsamples = DataTools.Subarray(recording.WavReader.Samples, startSample, subsegmentSampleCount);
var wr = new Acoustics.Tools.Wav.WavReader(subsamples, 1, 16, sampleRate);
subsegmentRecording = new AudioRecording(wr);
}
// INITIALISE a RESULTS STRUCTURE TO return
// initialize a result object in which to store SummaryIndexValues and SpectralIndexValues etc.
var result = new IndexCalculateResult(freqBinCount, indexProperties, indexCalculationDuration, subsegmentOffsetTimeSpan, config);
SummaryIndexValues summaryIndices = result.SummaryIndexValues;
SpectralIndexValues spectralIndices = result.SpectralIndexValues;
// set up default spectrogram to return
result.Sg = returnSonogramInfo ? GetSonogram(recording, windowSize: 1024) : null;
result.Hits = null;
result.TrackScores = new List <Plot>();
// ################################## FINSIHED SET-UP
// ################################## NOW GET THE AMPLITUDE SPECTORGRAMS
// EXTRACT ENVELOPE and SPECTROGRAM FROM SUBSEGMENT
var dspOutput1 = DSP_Frames.ExtractEnvelopeAndFfts(subsegmentRecording, frameSize, frameStep);
// Select band according to min and max bandwidth
int minBand = (int)(dspOutput1.AmplitudeSpectrogram.GetLength(1) * config.MinBandWidth);
int maxBand = (int)(dspOutput1.AmplitudeSpectrogram.GetLength(1) * config.MaxBandWidth) - 1;
dspOutput1.AmplitudeSpectrogram = MatrixTools.Submatrix(
dspOutput1.AmplitudeSpectrogram,
0,
minBand,
dspOutput1.AmplitudeSpectrogram.GetLength(0) - 1,
maxBand);
// TODO: Michael to review whether bandwidth filter should be moved to DSP_Frames??
// Recalculate NyquistBin and FreqBinWidth, because they change with band selection
//dspOutput1.NyquistBin = dspOutput1.AmplitudeSpectrogram.GetLength(1) - 1;
//dspOutput1.FreqBinWidth = sampleRate / (double)dspOutput1.AmplitudeSpectrogram.GetLength(1) / 2;
// Linear or Octave or Mel frequency scale? Set Linear as default.
var freqScale = new FrequencyScale(nyquist: nyquist, frameSize: frameSize, hertzGridInterval: 1000);
var freqScaleType = config.FrequencyScale;
bool octaveScale = freqScaleType == FreqScaleType.Linear125Octaves7Tones28Nyquist32000;
bool melScale = freqScaleType == FreqScaleType.Mel;
if (octaveScale)
{
// only allow one octave scale at the moment - for Jasco marine recordings.
// ASSUME fixed Occtave scale - USEFUL ONLY FOR JASCO 64000sr MARINE RECORDINGS
// If you wish to use other octave scale types then need to put in the config file and and set up recovery here.
freqScale = new FrequencyScale(FreqScaleType.Linear125Octaves7Tones28Nyquist32000);
// Recalculate the spectrogram according to octave scale. This option works only when have high SR recordings.
dspOutput1.AmplitudeSpectrogram = OctaveFreqScale.AmplitudeSpectra(
dspOutput1.AmplitudeSpectrogram,
dspOutput1.WindowPower,
sampleRate,
epsilon,
freqScale);
dspOutput1.NyquistBin = dspOutput1.AmplitudeSpectrogram.GetLength(1) - 1; // ASSUMPTION!!! Nyquist is in top Octave bin - not necessarily true!!
}
else if (melScale)
{
int minFreq = 0;
int maxFreq = recording.Nyquist;
dspOutput1.AmplitudeSpectrogram = MFCCStuff.MelFilterBank(
dspOutput1.AmplitudeSpectrogram,
config.MelScale,
recording.Nyquist,
minFreq,
maxFreq);
dspOutput1.NyquistBin = dspOutput1.AmplitudeSpectrogram.GetLength(1) - 1;
// TODO: This doesn't make any sense, since the frequency width changes for each bin. Probably need to set this to NaN.
// TODO: Whatever uses this value below, should probably be changed to not be depending on it.
dspOutput1.FreqBinWidth = sampleRate / (double)dspOutput1.AmplitudeSpectrogram.GetLength(1) / 2;
}
// NOW EXTRACT SIGNAL FOR BACKGROUND NOISE CALCULATION
// If the index calculation duration >= 30 seconds, then calculate BGN from the existing segment of recording.
bool doSeparateBgnNoiseCalculation = indexCalculationDuration.TotalSeconds + (2 * config.BgNoiseBuffer.TotalSeconds) < segmentDuration.TotalSeconds / 2;
var dspOutput2 = dspOutput1;
if (doSeparateBgnNoiseCalculation)
{
// GET a longer SUBSEGMENT FOR NOISE calculation with 5 sec buffer on either side.
// If the index calculation duration is shorter than 30 seconds, then need to calculate BGN noise from a longer length of recording
// i.e. need to add noiseBuffer either side. Typical noiseBuffer value = 5 seconds
int sampleBuffer = (int)(config.BgNoiseBuffer.TotalSeconds * sampleRate);
var bgnRecording = AudioRecording.GetRecordingSubsegment(recording, startSample, endSample, sampleBuffer);
// EXTRACT ENVELOPE and SPECTROGRAM FROM BACKGROUND NOISE SUBSEGMENT
dspOutput2 = DSP_Frames.ExtractEnvelopeAndFfts(bgnRecording, frameSize, frameStep);
// If necessary, recalculate the spectrogram according to octave scale. This option works only when have high SR recordings.
if (octaveScale)
{
// ASSUME fixed Occtave scale - USEFUL ONLY FOR JASCO 64000sr MARINE RECORDINGS
// If you wish to use other octave scale types then need to put in the config file and and set up recovery here.
dspOutput2.AmplitudeSpectrogram = OctaveFreqScale.AmplitudeSpectra(
dspOutput2.AmplitudeSpectrogram,
dspOutput2.WindowPower,
sampleRate,
epsilon,
freqScale);
dspOutput2.NyquistBin = dspOutput2.AmplitudeSpectrogram.GetLength(1) - 1; // ASSUMPTION!!! Nyquist is in top Octave bin - not necessarily true!!
}
}
// ###################################### BEGIN CALCULATION OF INDICES ##################################
// (A) ################################## EXTRACT SUMMARY INDICES FROM THE SIGNAL WAVEFORM ##################################
// average absolute value over the minute recording - not useful
// double[] avAbsolute = dspOutput1.Average;
double[] signalEnvelope = dspOutput1.Envelope;
double avgSignalEnvelope = signalEnvelope.Average();
// 10 times log of amplitude squared
summaryIndices.AvgSignalAmplitude = 20 * Math.Log10(avgSignalEnvelope);
// Deal with case where the signal waveform is continuous flat with values < 0.001. Has happened!!
// Although signal appears zero, this condition is required.
if (avgSignalEnvelope < 0.0001)
{
Logger.Debug("Segment skipped because avSignalEnvelope is < 0.001!");
summaryIndices.ZeroSignal = 1.0;
return(result);
}
// i. Check for clipping and high amplitude rates per second
summaryIndices.HighAmplitudeIndex = dspOutput1.HighAmplitudeCount / subsegmentDurationInSeconds;
summaryIndices.ClippingIndex = dspOutput1.ClipCount / subsegmentDurationInSeconds;
// ii. Calculate bg noise in dB
// Convert signal envelope to dB and subtract background noise. Default noise SD to calculate threshold = ZERO
double signalBgn = NoiseRemovalModal.CalculateBackgroundNoise(dspOutput2.Envelope);
summaryIndices.BackgroundNoise = signalBgn;
// iii: FRAME ENERGIES - convert signal to decibels and subtract background noise.
double[] dBEnvelope = SNR.Signal2Decibels(dspOutput1.Envelope);
double[] dBEnvelopeSansNoise = SNR.SubtractAndTruncate2Zero(dBEnvelope, signalBgn);
// iv: ACTIVITY for NOISE REDUCED SIGNAL ENVELOPE
// Calculate fraction of frames having acoustic activity
var activity = ActivityAndCover.CalculateActivity(dBEnvelopeSansNoise, frameStepTimeSpan);
summaryIndices.Activity = activity.FractionOfActiveFrames;
// v. average number of events per second whose duration > one frame
// average event duration in milliseconds - no longer calculated
//summaryIndices.AvgEventDuration = activity.avEventDuration;
summaryIndices.EventsPerSecond = activity.EventCount / subsegmentDurationInSeconds;
// vi. Calculate SNR and active frames SNR
summaryIndices.Snr = dBEnvelopeSansNoise.Max();
summaryIndices.AvgSnrOfActiveFrames = activity.ActiveAvDb;
// vii. ENTROPY of ENERGY ENVELOPE -- 1-Ht because want measure of concentration of acoustic energy.
double entropy = DataTools.EntropyNormalised(DataTools.SquareValues(signalEnvelope));
summaryIndices.TemporalEntropy = 1 - entropy;
// Note that the spectrogram has had the DC bin removed. i.e. has only 256 columns.
double[,] amplitudeSpectrogram = dspOutput1.AmplitudeSpectrogram; // get amplitude spectrogram.
// CALCULATE various NDSI (Normalised difference soundscape Index) FROM THE AMPLITUDE SPECTROGRAM
// These options proved to be highly correlated. Therefore only use tuple.Item 1 which derived from Power Spectral Density.
var tuple3 = SpectrogramTools.CalculateAvgSpectrumAndVarianceSpectrumFromAmplitudeSpectrogram(amplitudeSpectrogram);
summaryIndices.Ndsi = SpectrogramTools.CalculateNdsi(tuple3.Item1, sampleRate, 1000, 2000, 8000);
// (B) ################################## EXTRACT OSC SPECTRAL INDEX DIRECTLY FROM THE RECORDING ##################################
// Get the oscillation spectral index OSC separately from signal because need a different frame size etc.
var sampleLength = Oscillations2014.DefaultSampleLength;
var frameLength = Oscillations2014.DefaultFrameLength;
var sensitivity = Oscillations2014.DefaultSensitivityThreshold;
var spectralIndexShort = Oscillations2014.GetSpectralIndex_Osc(subsegmentRecording, frameLength, sampleLength, sensitivity);
// double length of the vector because want to work with 256 element vector for LDFC purposes
spectralIndices.OSC = DataTools.VectorDoubleLengthByAverageInterpolation(spectralIndexShort);
// (C) ################################## EXTRACT SPECTRAL INDICES FROM THE AMPLITUDE SPECTROGRAM ##################################
// i: CALCULATE SPECTRUM OF THE SUM OF FREQ BIN AMPLITUDES - used for later calculation of ACI
spectralIndices.SUM = MatrixTools.SumColumns(amplitudeSpectrogram);
// Calculate lower and upper boundary bin ids.
// Boundary between low & mid frequency bands is to avoid low freq bins containing anthropogenic noise. These biased index values away from biophony.
// Boundary of upper bird-band is to avoid high freq artefacts due to mp3.
int lowerBinBound = (int)Math.Ceiling(lowFreqBound / dspOutput1.FreqBinWidth);
int middleBinBound = (int)Math.Ceiling(midFreqBound / dspOutput1.FreqBinWidth);
// calculate number of freq bins in the bird-band.
int midBandBinCount = middleBinBound - lowerBinBound + 1;
if (octaveScale)
{
// the above frequency bin bounds do not apply with octave scale. Need to recalculate them suitable for Octave scale recording.
lowFreqBound = freqScale.LinearBound;
lowerBinBound = freqScale.GetBinIdForHerzValue(lowFreqBound);
midFreqBound = 8000; // This value appears suitable for Jasco Marine recordings. Not much happens above 8kHz.
//middleBinBound = freqScale.GetBinIdForHerzValue(midFreqBound);
middleBinBound = freqScale.GetBinIdInReducedSpectrogramForHerzValue(midFreqBound);
midBandBinCount = middleBinBound - lowerBinBound + 1;
}
// IFF there has been UP-SAMPLING, calculate bin of the original audio nyquist. this will be less than SR/2.
// original sample rate can be anything 11.0-44.1 kHz.
int originalNyquist = sampleRateOfOriginalAudioFile / 2;
// if upsampling has been done
if (dspOutput1.NyquistFreq > originalNyquist)
{
dspOutput1.NyquistFreq = originalNyquist;
dspOutput1.NyquistBin = (int)Math.Floor(originalNyquist / dspOutput1.FreqBinWidth); // note that binwidth does not change
}
// ii: CALCULATE THE ACOUSTIC COMPLEXITY INDEX
spectralIndices.DIF = AcousticComplexityIndex.SumOfAmplitudeDifferences(amplitudeSpectrogram);
double[] aciSpectrum = AcousticComplexityIndex.CalculateAci(amplitudeSpectrogram);
spectralIndices.ACI = aciSpectrum;
// remove low freq band of ACI spectrum and store average ACI value
double[] reducedAciSpectrum = DataTools.Subarray(aciSpectrum, lowerBinBound, midBandBinCount);
summaryIndices.AcousticComplexity = reducedAciSpectrum.Average();
// iii: CALCULATE the H(t) or Temporal ENTROPY Spectrum and then reverse the values i.e. calculate 1-Ht for energy concentration
double[] temporalEntropySpectrum = AcousticEntropy.CalculateTemporalEntropySpectrum(amplitudeSpectrogram);
for (int i = 0; i < temporalEntropySpectrum.Length; i++)
{
temporalEntropySpectrum[i] = 1 - temporalEntropySpectrum[i];
}
spectralIndices.ENT = temporalEntropySpectrum;
// iv: remove background noise from the amplitude spectrogram
// First calculate the noise profile from the amplitude sepctrogram
double[] spectralAmplitudeBgn = NoiseProfile.CalculateBackgroundNoise(dspOutput2.AmplitudeSpectrogram);
amplitudeSpectrogram = SNR.TruncateBgNoiseFromSpectrogram(amplitudeSpectrogram, spectralAmplitudeBgn);
// AMPLITUDE THRESHOLD for smoothing background, nhThreshold, assumes background noise ranges around -40dB.
// This value corresponds to approximately 6dB above backgorund.
amplitudeSpectrogram = SNR.RemoveNeighbourhoodBackgroundNoise(amplitudeSpectrogram, nhThreshold: 0.015);
////ImageTools.DrawMatrix(spectrogramData, @"C:\SensorNetworks\WavFiles\Crows\image.png", false);
////DataTools.writeBarGraph(modalValues);
result.AmplitudeSpectrogram = amplitudeSpectrogram;
// v: ENTROPY OF AVERAGE SPECTRUM & VARIANCE SPECTRUM - at this point the spectrogram is a noise reduced amplitude spectrogram
var tuple = AcousticEntropy.CalculateSpectralEntropies(amplitudeSpectrogram, lowerBinBound, midBandBinCount);
// ENTROPY of spectral averages - Reverse the values i.e. calculate 1-Hs and 1-Hv, and 1-Hcov for energy concentration
summaryIndices.EntropyOfAverageSpectrum = 1 - tuple.Item1;
// ENTROPY of spectrum of Variance values
summaryIndices.EntropyOfVarianceSpectrum = 1 - tuple.Item2;
// ENTROPY of spectrum of Coefficient of Variation values
summaryIndices.EntropyOfCoVSpectrum = 1 - tuple.Item3;
// vi: ENTROPY OF DISTRIBUTION of maximum SPECTRAL PEAKS.
// First extract High band SPECTROGRAM which is now noise reduced
double entropyOfPeaksSpectrum = AcousticEntropy.CalculateEntropyOfSpectralPeaks(amplitudeSpectrogram, lowerBinBound, middleBinBound);
summaryIndices.EntropyOfPeaksSpectrum = 1 - entropyOfPeaksSpectrum;
// ######################################################################################################################################################
// (C) ################################## EXTRACT SPECTRAL INDICES FROM THE DECIBEL SPECTROGRAM ##################################
// i: Convert amplitude spectrogram to deciBels and calculate the dB background noise profile
double[,] deciBelSpectrogram = MFCCStuff.DecibelSpectra(dspOutput2.AmplitudeSpectrogram, dspOutput2.WindowPower, sampleRate, epsilon);
double[] spectralDecibelBgn = NoiseProfile.CalculateBackgroundNoise(deciBelSpectrogram);
spectralIndices.BGN = spectralDecibelBgn;
// ii: Calculate the noise reduced decibel spectrogram derived from segment recording.
// REUSE the var decibelSpectrogram but this time using dspOutput1.
deciBelSpectrogram = MFCCStuff.DecibelSpectra(dspOutput1.AmplitudeSpectrogram, dspOutput1.WindowPower, sampleRate, epsilon);
deciBelSpectrogram = SNR.TruncateBgNoiseFromSpectrogram(deciBelSpectrogram, spectralDecibelBgn);
deciBelSpectrogram = SNR.RemoveNeighbourhoodBackgroundNoise(deciBelSpectrogram, nhThreshold: 2.0);
// iii: CALCULATE noise reduced AVERAGE DECIBEL SPECTRUM
spectralIndices.PMN = SpectrogramTools.CalculateAvgDecibelSpectrumFromDecibelSpectrogram(deciBelSpectrogram);
// iv: CALCULATE SPECTRAL COVER.
// NOTE: at this point, decibelSpectrogram is noise reduced. All values >= 0.0
// FreqBinWidth can be accessed, if required, through dspOutput1.FreqBinWidth
double dBThreshold = ActivityAndCover.DefaultActivityThresholdDb; // dB THRESHOLD for calculating spectral coverage
var spActivity = ActivityAndCover.CalculateSpectralEvents(deciBelSpectrogram, dBThreshold, frameStepTimeSpan, lowerBinBound, middleBinBound);
spectralIndices.CVR = spActivity.CoverSpectrum;
spectralIndices.EVN = spActivity.EventSpectrum;
summaryIndices.HighFreqCover = spActivity.HighFreqBandCover;
summaryIndices.MidFreqCover = spActivity.MidFreqBandCover;
summaryIndices.LowFreqCover = spActivity.LowFreqBandCover;
// ######################################################################################################################################################
// v: CALCULATE SPECTRAL PEAK TRACKS and RIDGE indices.
// NOTE: at this point, the var decibelSpectrogram is noise reduced. i.e. all its values >= 0.0
// Detecting ridges or spectral peak tracks requires using a 5x5 mask which has edge effects.
// This becomes significant if we have a short indexCalculationDuration.
// Consequently if the indexCalculationDuration < 10 seconds then we revert back to the recording and cut out a recording segment that includes
// a buffer for edge effects. In most cases however, we can just use the decibel spectrogram already calculated and ignore the edge effects.
double peakThreshold = 6.0; //dB
SpectralPeakTracks sptInfo;
if (indexCalculationDuration.TotalSeconds < 10.0)
{
// calculate a new decibel spectrogram
sptInfo = SpectralPeakTracks.CalculateSpectralPeakTracks(recording, startSample, endSample, frameSize, octaveScale, peakThreshold);
}
else
{
// use existing decibel spectrogram
sptInfo = new SpectralPeakTracks(deciBelSpectrogram, peakThreshold);
}
spectralIndices.SPT = sptInfo.SptSpectrum;
spectralIndices.RHZ = sptInfo.RhzSpectrum;
spectralIndices.RVT = sptInfo.RvtSpectrum;
spectralIndices.RPS = sptInfo.RpsSpectrum;
spectralIndices.RNG = sptInfo.RngSpectrum;
summaryIndices.SptDensity = sptInfo.TrackDensity;
// these are two other indices that I tried but they do not seem to add anything of interest.
//summaryIndices.AvgSptDuration = sptInfo.AvTrackDuration;
//summaryIndices.SptPerSecond = sptInfo.TotalTrackCount / subsegmentSecondsDuration;
// ######################################################################################################################################################
// vi: CLUSTERING - FIRST DETERMINE IF IT IS WORTH DOING
// return if (activeFrameCount too small || eventCount == 0 || short index calc duration) because no point doing clustering
if (activity.ActiveFrameCount <= 2 || Math.Abs(activity.EventCount) < 0.01 || indexCalculationDuration.TotalSeconds < 15)
{
// IN ADDITION return if indexCalculationDuration < 15 seconds because no point doing clustering on short time segment
// NOTE: Activity was calculated with 3dB threshold AFTER backgroundnoise removal.
//summaryIndices.AvgClusterDuration = TimeSpan.Zero;
summaryIndices.ClusterCount = 0;
summaryIndices.ThreeGramCount = 0;
return(result);
}
// YES WE WILL DO CLUSTERING! to determine cluster count (spectral diversity) and spectral persistence.
// Only use midband decibel SPECTRUM. In June 2016, the mid-band (i.e. the bird-band) was set to lowerBound=1000Hz, upperBound=8000hz.
// Actually do clustering of binary spectra. Must first threshold
double binaryThreshold = SpectralClustering.DefaultBinaryThresholdInDecibels;
var midBandSpectrogram = MatrixTools.Submatrix(deciBelSpectrogram, 0, lowerBinBound, deciBelSpectrogram.GetLength(0) - 1, middleBinBound);
var clusterInfo = SpectralClustering.ClusterTheSpectra(midBandSpectrogram, lowerBinBound, middleBinBound, binaryThreshold);
// Store two summary index values from cluster info
summaryIndices.ClusterCount = clusterInfo.ClusterCount;
summaryIndices.ThreeGramCount = clusterInfo.TriGramUniqueCount;
// As of May 2017, no longer store clustering results superimposed on spectrogram.
// If you want to see this, then call the TEST methods in class SpectralClustering.cs.
// #######################################################################################################################################################
// vii: set up other info to return
var freqPeaks = SpectralPeakTracks.ConvertSpectralPeaksToNormalisedArray(deciBelSpectrogram);
var scores = new List <Plot>
{
new Plot("Decibels", DataTools.normalise(dBEnvelopeSansNoise), ActivityAndCover.DefaultActivityThresholdDb),
new Plot("Active Frames", DataTools.Bool2Binary(activity.ActiveFrames), 0.0),
new Plot("Max Frequency", freqPeaks, 0.0), // relative location of freq maxima in spectra
};
result.Hits = sptInfo.Peaks;
result.TrackScores = scores;
return(result);
} // end Calculation of Summary and Spectral Indices
