/// <summary>
/// A METHOD TO DETECT HARMONICS IN THE sub-band of a spectrogram.
/// This method assume the matrix is derived from a spectrogram rotated so that the matrix rows are spectral columns of the spectrogram.
/// Developed for GenericRecognizer of harmonics.
/// WARNING: As of March 2020, this method averages the values in five adjacent frames. This is to reduce noise.
/// But it requires that the frequency of any potential formants is not changing rapidly.
/// THis may not be suitable for detecting human speech. However can reduce the frame step.
/// </summary>
/// <param name="m">spectrogram data matrix.</param>
/// <param name="dBThreshold">Minimum sound level.</param>
/// <returns>three arrays: dBArray, intensity, maxIndexArray.</returns>
public static Tuple <double[], double[], int[]> DetectHarmonicsInSpectrogramData(double[,] m, double dBThreshold)
{
int rowCount = m.GetLength(0);
int colCount = m.GetLength(1);
var binCount = m.GetLength(1);
//set up the cosine coefficients
double[,] cosines = MFCCStuff.Cosines(binCount, binCount);
// set up arrays to store decibels, formant intensity and max index.
var dBArray = new double[rowCount];
var intensity = new double[rowCount];
var maxIndexArray = new int[rowCount];
// for all time frames
for (int t = 2; t < rowCount - 2; t++)
{
// get average of five adjacent frames
var frame1 = MatrixTools.GetRow(m, t - 2);
var frame2 = MatrixTools.GetRow(m, t - 1);
var frame3 = MatrixTools.GetRow(m, t);
var frame4 = MatrixTools.GetRow(m, t + 1);
var frame5 = MatrixTools.GetRow(m, t + 2);
var frame = new double[colCount];
for (int i = 0; i < colCount; i++)
{
frame[i] = (frame1[i] + frame2[i] + frame3[i] + frame4[i] + frame5[i]) / 5;
}
double maxValue = frame.Max();
dBArray[t] = maxValue;
if (maxValue < dBThreshold)
{
continue;
}
double[] xr = AutoAndCrossCorrelation.AutoCrossCorr(frame);
// xr has twice length of frame and is symmetrical.
// Require only first half.
double[] normXr = new double[colCount];
for (int i = 0; i < colCount; i++)
{
// Would normally normalise the xcorr values for overlap count.
// But for harmonics, this introduces too much noise - need to give less weight to the less overlapped values.
//normXr[i] = xr[i] / (colCount - i);
normXr[i] = xr[i];
}
// now do DCT across the auto cross xr
int lowerDctBound = 2;
var dctCoefficients = Oscillations2012.DoDct(normXr, cosines, lowerDctBound);
int indexOfMaxValue = DataTools.GetMaxIndex(dctCoefficients);
intensity[t] = dctCoefficients[indexOfMaxValue];
maxIndexArray[t] = indexOfMaxValue;
}
return(Tuple.Create(dBArray, intensity, maxIndexArray));
}
/// <summary>
/// A METHOD TO DETECT HARMONICS IN THE sub-band of a sonogram.
/// This method assume the matrix is derived from a spectrogram rotated so that the matrix rows are spectral columns of sonogram.
/// Developed for GenericRecognizer of harmonics.
/// </summary>
/// <param name="m">data matrix.</param>
/// <param name="dBThreshold">Minimum sound level.</param>
/// <returns>two arrays.</returns>
public static Tuple <double[], double[], int[]> DetectHarmonicsInSonogramMatrix(double[,] m, double dBThreshold)
{
int rowCount = m.GetLength(0);
int colCount = m.GetLength(1);
double[] dBArray = new double[rowCount];
var intensity = new double[rowCount]; //an array of formant intensity
var maxIndexArray = new int[rowCount]; //an array of max value index values
var binCount = m.GetLength(1);
double[,] cosines = MFCCStuff.Cosines(binCount, binCount); //set up the cosine coefficients
// for all time frames
for (int t = 0; t < rowCount; t++)
{
var frame = MatrixTools.GetRow(m, t);
double maxValue = frame.Max();
dBArray[t] = maxValue;
if (maxValue < dBThreshold)
{
continue;
}
double[] xr = AutoAndCrossCorrelation.AutoCrossCorr(frame);
// xr has twice length of frame and is symmetrical.
// Require only first half. Also need to normalise the values for overlap count.
double[] normXr = new double[colCount];
for (int i = 0; i < colCount; i++)
{
normXr[i] = xr[i] / (colCount - i);
}
// now do DCT across the auto cross xr
int lowerDctBound = 2;
var dctCoefficients = Oscillations2012.DoDct(normXr, cosines, lowerDctBound);
int indexOfMaxValue = DataTools.GetMaxIndex(dctCoefficients);
intensity[t] = dctCoefficients[indexOfMaxValue];
maxIndexArray[t] = indexOfMaxValue;
} // frames = rows of matrix
return(Tuple.Create(dBArray, intensity, maxIndexArray));
}
/// <summary>
/// ################ THE KEY ANALYSIS METHOD for TRILLS
///
/// See Anthony's ExempliGratia.Recognize() method in order to see how to use methods for config profiles.
/// </summary>
/// <param name="recording"></param>
/// <param name="sonoConfig"></param>
/// <param name="lwConfig"></param>
/// <param name="returnDebugImage"></param>
/// <param name="segmentStartOffset"></param>
/// <returns></returns>
private static Tuple <BaseSonogram, double[, ], double[], List <AcousticEvent>, Image> Analysis(
AudioRecording recording,
SonogramConfig sonoConfig,
LitoriaWatjulumConfig lwConfig,
bool returnDebugImage,
TimeSpan segmentStartOffset)
{
double intensityThreshold = lwConfig.IntensityThreshold;
double minDuration = lwConfig.MinDurationOfTrill; // seconds
double maxDuration = lwConfig.MaxDurationOfTrill; // seconds
double minPeriod = lwConfig.MinPeriod; // seconds
double maxPeriod = lwConfig.MaxPeriod; // seconds
if (recording == null)
{
LoggedConsole.WriteLine("AudioRecording == null. Analysis not possible.");
return(null);
}
//i: MAKE SONOGRAM
//TimeSpan tsRecordingtDuration = recording.Duration();
int sr = recording.SampleRate;
double freqBinWidth = sr / (double)sonoConfig.WindowSize;
double framesPerSecond = freqBinWidth;
// duration of DCT in seconds - want it to be about 3X or 4X the expected maximum period
double dctDuration = 4 * maxPeriod;
// duration of DCT in frames
int dctLength = (int)Math.Round(framesPerSecond * dctDuration);
// set up the cosine coefficients
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength);
int upperBandMinBin = (int)Math.Round(lwConfig.UpperBandMinHz / freqBinWidth) + 1;
int upperBandMaxBin = (int)Math.Round(lwConfig.UpperBandMaxHz / freqBinWidth) + 1;
int lowerBandMinBin = (int)Math.Round(lwConfig.LowerBandMinHz / freqBinWidth) + 1;
int lowerBandMaxBin = (int)Math.Round(lwConfig.LowerBandMaxHz / freqBinWidth) + 1;
BaseSonogram sonogram = new SpectrogramStandard(sonoConfig, recording.WavReader);
int rowCount = sonogram.Data.GetLength(0);
//int colCount = sonogram.Data.GetLength(1);
double[] lowerArray = MatrixTools.GetRowAveragesOfSubmatrix(sonogram.Data, 0, lowerBandMinBin, rowCount - 1, lowerBandMaxBin);
double[] upperArray = MatrixTools.GetRowAveragesOfSubmatrix(sonogram.Data, 0, upperBandMinBin, rowCount - 1, upperBandMaxBin);
//lowerArray = DataTools.filterMovingAverage(lowerArray, 3);
//upperArray = DataTools.filterMovingAverage(upperArray, 3);
double[] amplitudeScores = DataTools.SumMinusDifference(lowerArray, upperArray);
double[] differenceScores = DspFilters.SubtractBaseline(amplitudeScores, 7);
// Could smooth here rather than above. Above seemed slightly better?
//amplitudeScores = DataTools.filterMovingAverage(amplitudeScores, 7);
//differenceScores = DataTools.filterMovingAverage(differenceScores, 7);
//iii: CONVERT decibel sum-diff SCORES TO ACOUSTIC TRILL EVENTS
var predictedTrillEvents = AcousticEvent.ConvertScoreArray2Events(
amplitudeScores,
lwConfig.LowerBandMinHz,
lwConfig.UpperBandMaxHz,
sonogram.FramesPerSecond,
freqBinWidth,
lwConfig.DecibelThreshold,
minDuration,
maxDuration,
segmentStartOffset);
for (int i = 0; i < differenceScores.Length; i++)
{
if (differenceScores[i] < 1.0)
{
differenceScores[i] = 0.0;
}
}
// LOOK FOR TRILL EVENTS
// init the score array
double[] scores = new double[rowCount];
// var hits = new double[rowCount, colCount];
double[,] hits = null;
// init confirmed events
var confirmedEvents = new List <AcousticEvent>();
// add names into the returned events
foreach (var ae in predictedTrillEvents)
{
int eventStart = ae.Oblong.RowTop;
int eventWidth = ae.Oblong.RowWidth;
int step = 2;
double maximumIntensity = 0.0;
// scan the event to get oscillation period and intensity
for (int i = eventStart - (dctLength / 2); i < eventStart + eventWidth - (dctLength / 2); i += step)
{
// Look for oscillations in the difference array
double[] differenceArray = DataTools.Subarray(differenceScores, i, dctLength);
double oscilFreq;
double period;
double intensity;
Oscillations2014.GetOscillation(differenceArray, framesPerSecond, cosines, out oscilFreq, out period, out intensity);
bool periodWithinBounds = period > minPeriod && period < maxPeriod;
//Console.WriteLine($"step={i} period={period:f4}");
if (!periodWithinBounds)
{
continue;
}
for (int j = 0; j < dctLength; j++) //lay down score for sample length
{
if (scores[i + j] < intensity)
{
scores[i + j] = intensity;
}
}
if (maximumIntensity < intensity)
{
maximumIntensity = intensity;
}
}
// add abbreviatedSpeciesName into event
if (maximumIntensity >= intensityThreshold)
{
ae.Name = $"{lwConfig.AbbreviatedSpeciesName}.{lwConfig.ProfileNames[0]}";
ae.Score_MaxInEvent = maximumIntensity;
ae.Profile = lwConfig.ProfileNames[0];
confirmedEvents.Add(ae);
}
}
//######################################################################
// LOOK FOR TINK EVENTS
// CONVERT decibel sum-diff SCORES TO ACOUSTIC EVENTS
double minDurationOfTink = lwConfig.MinDurationOfTink; // seconds
double maxDurationOfTink = lwConfig.MaxDurationOfTink; // seconds
// want stronger threshold for tink because brief.
double tinkDecibelThreshold = lwConfig.DecibelThreshold + 3.0;
var predictedTinkEvents = AcousticEvent.ConvertScoreArray2Events(
amplitudeScores,
lwConfig.LowerBandMinHz,
lwConfig.UpperBandMaxHz,
sonogram.FramesPerSecond,
freqBinWidth,
tinkDecibelThreshold,
minDurationOfTink,
maxDurationOfTink,
segmentStartOffset);
foreach (var ae2 in predictedTinkEvents)
{
// Prune the list of potential acoustic events, for example using Cosine Similarity.
//rowtop, rowWidth
//int eventStart = ae2.Oblong.RowTop;
//int eventWidth = ae2.Oblong.RowWidth;
//int step = 2;
//double maximumIntensity = 0.0;
// add abbreviatedSpeciesName into event
//if (maximumIntensity >= intensityThreshold)
//{
ae2.Name = $"{lwConfig.AbbreviatedSpeciesName}.{lwConfig.ProfileNames[1]}";
//ae2.Score_MaxInEvent = maximumIntensity;
ae2.Profile = lwConfig.ProfileNames[1];
confirmedEvents.Add(ae2);
//}
}
//######################################################################
var scorePlot = new Plot(lwConfig.SpeciesName, scores, intensityThreshold);
Image debugImage = null;
if (returnDebugImage)
{
// display a variety of debug score arrays
double[] normalisedScores;
double normalisedThreshold;
DataTools.Normalise(amplitudeScores, lwConfig.DecibelThreshold, out normalisedScores, out normalisedThreshold);
var sumDiffPlot = new Plot("Sum Minus Difference", normalisedScores, normalisedThreshold);
DataTools.Normalise(differenceScores, lwConfig.DecibelThreshold, out normalisedScores, out normalisedThreshold);
var differencePlot = new Plot("Baseline Removed", normalisedScores, normalisedThreshold);
var debugPlots = new List <Plot> {
scorePlot, sumDiffPlot, differencePlot
};
debugImage = DrawDebugImage(sonogram, confirmedEvents, debugPlots, hits);
}
// return new sonogram because it makes for more easy interpretation of the image
var returnSonoConfig = new SonogramConfig
{
SourceFName = recording.BaseName,
WindowSize = 512,
WindowOverlap = 0,
// the default window is HAMMING
//WindowFunction = WindowFunctions.HANNING.ToString(),
//WindowFunction = WindowFunctions.NONE.ToString(),
// if do not use noise reduction can get a more sensitive recogniser.
//NoiseReductionType = NoiseReductionType.NONE,
NoiseReductionType = SNR.KeyToNoiseReductionType("STANDARD"),
};
BaseSonogram returnSonogram = new SpectrogramStandard(returnSonoConfig, recording.WavReader);
return(Tuple.Create(returnSonogram, hits, scores, confirmedEvents, debugImage));
} //Analysis()
public static Tuple <double[]> Execute_MFCC_XCOR(double[,] target, double dynamicRange, SpectrogramStandard sonogram,
List <AcousticEvent> segments, int minHz, int maxHz, double minDuration)
{
Log.WriteLine("SEARCHING FOR EVENTS LIKE TARGET.");
if (segments == null)
{
return(null);
}
int minBin = (int)(minHz / sonogram.FBinWidth);
int maxBin = (int)(maxHz / sonogram.FBinWidth);
int targetLength = target.GetLength(0);
//set up the matrix of cosine coefficients
int coeffCount = 12; //only use first 12 coefficients.
int binCount = target.GetLength(1); //number of filters in filter bank
double[,] cosines = MFCCStuff.Cosines(binCount, coeffCount + 1); //set up the cosine coefficients
//adjust target's dynamic range to that set by user
target = SNR.SetDynamicRange(target, 3.0, dynamicRange); //set event's dynamic range
target = MFCCStuff.Cepstra(target, coeffCount, cosines);
double[] v1 = DataTools.Matrix2Array(target);
v1 = DataTools.normalise2UnitLength(v1);
string imagePath2 = @"C:\SensorNetworks\Output\FELT_Currawong\target.png";
var result1 = BaseSonogram.Data2ImageData(target);
var image = result1.Item1;
ImageTools.DrawMatrix(image, 1, 1, imagePath2);
double[] scores = new double[sonogram.FrameCount];
foreach (AcousticEvent av in segments)
{
Log.WriteLine("SEARCHING SEGMENT.");
int startRow = (int)Math.Round(av.TimeStart * sonogram.FramesPerSecond);
int endRow = (int)Math.Round(av.TimeEnd * sonogram.FramesPerSecond);
if (endRow >= sonogram.FrameCount)
{
endRow = sonogram.FrameCount - 1;
}
endRow -= targetLength;
if (endRow <= startRow)
{
endRow = startRow + 1; //want minimum of one row
}
for (int r = startRow; r < endRow; r++)
{
double[,] matrix = DataTools.Submatrix(sonogram.Data, r, minBin, r + targetLength - 1, maxBin);
matrix = SNR.SetDynamicRange(matrix, 3.0, dynamicRange); //set event's dynamic range
//string imagePath2 = @"C:\SensorNetworks\Output\FELT_Gecko\compare.png";
//var image = BaseSonogram.Data2ImageData(matrix);
//ImageTools.DrawMatrix(image, 1, 1, imagePath2);
matrix = MFCCStuff.Cepstra(matrix, coeffCount, cosines);
double[] v2 = DataTools.Matrix2Array(matrix);
v2 = DataTools.normalise2UnitLength(v2);
double crossCor = DataTools.DotProduct(v1, v2);
scores[r] = crossCor;
} //end of rows in segment
} //foreach (AcousticEvent av in segments)
var tuple = Tuple.Create(scores);
return(tuple);
}
/// <summary>
/// THIS METHOD NO LONGER IN USE.
/// NOT USEFUL FOR ANIMAL CALLS.
/// Tried this but it is suitable only when there is guarantee of numerous spectral tracks as in the vowels of human speech.
/// It yields SPURIOUS RESULTS where there is only one whistle track.
/// </summary>
public static double[,] DetectHarmonicsUsingDCT(double[,] matrix, int minBin, int maxBin, int hzWidth, bool normaliseDCT, int minPeriod, int maxPeriod, double dctThreshold)
{
int dctLength = maxBin - minBin + 1; //DCT spans N freq bins
int minIndex = (int)(hzWidth / (double)maxPeriod * 2); //Times 0.5 because index = Pi and not 2Pi
int maxIndex = (int)(hzWidth / (double)minPeriod * 2); //Times 0.5 because index = Pi and not 2Pi
//double period = hzWidth / (double)indexOfMaxValue * 2; //Times 2 because index = Pi and not 2Pi
if (maxIndex > dctLength)
{
maxIndex = dctLength; //safety check in case of future changes to code.
}
int rows = matrix.GetLength(0);
int cols = matrix.GetLength(1);
double[,] hits = new double[rows, cols];
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
for (int r = 0; r < rows - dctLength; r++)
{
//for (int c = minBin; c <= minBin; c++)//traverse columns - skip DC column
//{
var array = new double[dctLength];
//accumulate J rows of values
for (int i = 0; i < dctLength; i++)
{
for (int j = 0; j < 5; j++)
{
array[i] += matrix[r + j, minBin + i];
}
}
array = DataTools.SubtractMean(array);
// DataTools.writeBarGraph(array);
double[] dct = MFCCStuff.DCT(array, cosines);
for (int i = 0; i < dctLength; i++)
{
dct[i] = Math.Abs(dct[i]); //convert to absolute values
}
for (int i = 0; i < 5; i++)
{
dct[i] = 0.0; //remove low freq values from consideration
}
if (normaliseDCT)
{
dct = DataTools.normalise2UnitLength(dct);
}
int indexOfMaxValue = DataTools.GetMaxIndex(dct);
//DataTools.writeBarGraph(dct);
double period = hzWidth / (double)indexOfMaxValue * 2; //Times 2 because index = Pi and not 2Pi
//mark DCT location with harmonic freq, only if harmonic freq is in correct range and amplitude
if (indexOfMaxValue >= minIndex && indexOfMaxValue <= maxIndex && dct[indexOfMaxValue] > dctThreshold)
{
for (int i = 0; i < dctLength; i++)
{
hits[r, minBin + i] = period;
}
for (int i = 0; i < dctLength; i++)
{
hits[r + 1, minBin + i] = period; //alternate row
}
}
//c += 5; //skip columns
//}
r++; //do alternate row
}
return(hits);
}
/// <summary>
/// Currently this method is called by only one species recognizer - LitoriaCaerulea.
/// </summary>
/// <param name="ipArray">an array of decibel values.</param>
/// <param name="framesPerSecond">the frame rate.</param>
/// <param name="decibelThreshold">Ignore frames below this threshold.</param>
/// <param name="dctDuration">Duration in seconds of the required DCT.</param>
/// <param name="minOscFreq">minimum oscillation frequency.</param>
/// <param name="maxOscFreq">maximum oscillation frequency.</param>
/// <param name="dctThreshold">Threshold for the maximum DCT coefficient.</param>
/// <param name="dctScores">an array of dct scores.</param>
/// <param name="oscFreq">an array of oscillation frequencies.</param>
public static void DetectOscillations(
double[] ipArray,
double framesPerSecond,
double decibelThreshold,
double dctDuration,
double minOscFreq,
double maxOscFreq,
double dctThreshold,
out double[] dctScores,
out double[] oscFreq)
{
int dctLength = (int)Math.Round(framesPerSecond * dctDuration);
int minIndex = (int)(minOscFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int maxIndex = (int)(maxOscFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
if (maxIndex > dctLength)
{
LoggedConsole.WriteWarnLine("MaxIndex > DCT length. Therefore set maxIndex = DCT length.");
maxIndex = dctLength;
}
int length = ipArray.Length;
dctScores = new double[length];
oscFreq = new double[length];
//set up the cosine coefficients
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength);
//following two lines write bmp image of cosine matrix values for checking.
//string bmpPath = @"C:\SensorNetworks\Output\cosines.png";
//ImageTools.DrawMatrix(cosines, bmpPath, true);
for (int r = 1; r < length - dctLength; r++)
{
// only stop if current location is a peak
if (ipArray[r] < ipArray[r - 1] || ipArray[r] < ipArray[r + 1])
{
continue;
}
// only stop if current location is a peak
if (ipArray[r] < decibelThreshold)
{
continue;
}
// extract array and ready for DCT
var dctArray = DataTools.Subarray(ipArray, r, dctLength);
dctArray = DataTools.SubtractMean(dctArray);
double[] dctCoefficient = MFCCStuff.DCT(dctArray, cosines);
// convert to absolute values because not interested in negative values due to phase.
for (int i = 0; i < dctLength; i++)
{
dctCoefficient[i] = Math.Abs(dctCoefficient[i]);
}
// remove low freq oscillations from consideration
int thresholdIndex = minIndex / 4;
for (int i = 0; i < thresholdIndex; i++)
{
dctCoefficient[i] = 0.0;
}
dctCoefficient = DataTools.normalise2UnitLength(dctCoefficient);
int indexOfMaxValue = DataTools.GetMaxIndex(dctCoefficient);
//mark DCT location with oscillation freq, only if oscillation freq is in correct range and amplitude
if (indexOfMaxValue >= minIndex && indexOfMaxValue <= maxIndex && dctCoefficient[indexOfMaxValue] > dctThreshold)
{
for (int i = 0; i < dctLength; i++)
{
if (dctScores[r + i] < dctCoefficient[indexOfMaxValue])
{
dctScores[r + i] = dctCoefficient[indexOfMaxValue];
oscFreq[r + i] = indexOfMaxValue / dctDuration / 2;
}
}
}
}
}
/// <summary>
/// THE KEY ANALYSIS METHOD.
/// </summary>
public static Tuple <BaseSonogram, double[, ], double[], List <AcousticEvent>, Image> Analysis(
AudioRecording recording,
SonogramConfig sonoConfig,
LitoriaBicolorConfig lbConfig,
bool drawDebugImage,
TimeSpan segmentStartOffset)
{
double decibelThreshold = lbConfig.DecibelThreshold; //dB
double intensityThreshold = lbConfig.IntensityThreshold;
//double eventThreshold = lbConfig.EventThreshold; //in 0-1
if (recording == null)
{
LoggedConsole.WriteLine("AudioRecording == null. Analysis not possible.");
return(null);
}
//i: MAKE SONOGRAM
//TimeSpan tsRecordingtDuration = recording.Duration();
int sr = recording.SampleRate;
double freqBinWidth = sr / (double)sonoConfig.WindowSize;
double framesPerSecond = freqBinWidth;
// duration of DCT in seconds - want it to be about 3X or 4X the expected maximum period
double dctDuration = 3 * lbConfig.MaxPeriod;
// duration of DCT in frames
int dctLength = (int)Math.Round(framesPerSecond * dctDuration);
// set up the cosine coefficients
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength);
int upperBandMinBin = (int)Math.Round(lbConfig.UpperBandMinHz / freqBinWidth) + 1;
int upperBandMaxBin = (int)Math.Round(lbConfig.UpperBandMaxHz / freqBinWidth) + 1;
int lowerBandMinBin = (int)Math.Round(lbConfig.LowerBandMinHz / freqBinWidth) + 1;
int lowerBandMaxBin = (int)Math.Round(lbConfig.LowerBandMaxHz / freqBinWidth) + 1;
BaseSonogram sonogram = new SpectrogramStandard(sonoConfig, recording.WavReader);
int rowCount = sonogram.Data.GetLength(0);
int colCount = sonogram.Data.GetLength(1);
double[] lowerArray = MatrixTools.GetRowAveragesOfSubmatrix(sonogram.Data, 0, lowerBandMinBin, rowCount - 1, lowerBandMaxBin);
double[] upperArray = MatrixTools.GetRowAveragesOfSubmatrix(sonogram.Data, 0, upperBandMinBin, rowCount - 1, upperBandMaxBin);
//lowerArray = DataTools.filterMovingAverage(lowerArray, 3);
//upperArray = DataTools.filterMovingAverage(upperArray, 3);
double[] amplitudeScores = DataTools.SumMinusDifference(lowerArray, upperArray);
double[] differenceScores = DspFilters.PreEmphasis(amplitudeScores, 1.0);
// Could smooth here rather than above. Above seemed slightly better?
amplitudeScores = DataTools.filterMovingAverage(amplitudeScores, 7);
differenceScores = DataTools.filterMovingAverage(differenceScores, 7);
//iii: CONVERT decibel sum-diff SCORES TO ACOUSTIC EVENTS
var predictedEvents = AcousticEvent.ConvertScoreArray2Events(
amplitudeScores,
lbConfig.LowerBandMinHz,
lbConfig.UpperBandMaxHz,
sonogram.FramesPerSecond,
freqBinWidth,
decibelThreshold,
lbConfig.MinDuration,
lbConfig.MaxDuration,
segmentStartOffset);
for (int i = 0; i < differenceScores.Length; i++)
{
if (differenceScores[i] < 1.0)
{
differenceScores[i] = 0.0;
}
}
// init the score array
double[] scores = new double[rowCount];
//iii: CONVERT SCORES TO ACOUSTIC EVENTS
// var hits = new double[rowCount, colCount];
double[,] hits = null;
// init confirmed events
var confirmedEvents = new List <AcousticEvent>();
// add names into the returned events
foreach (var ae in predictedEvents)
{
//rowtop, rowWidth
int eventStart = ae.Oblong.RowTop;
int eventWidth = ae.Oblong.RowWidth;
int step = 2;
double maximumIntensity = 0.0;
// scan the event to get oscillation period and intensity
for (int i = eventStart - (dctLength / 2); i < eventStart + eventWidth - (dctLength / 2); i += step)
{
// Look for oscillations in the difference array
double[] differenceArray = DataTools.Subarray(differenceScores, i, dctLength);
Oscillations2014.GetOscillationUsingDct(differenceArray, framesPerSecond, cosines, out var oscilFreq, out var period, out var intensity);
bool periodWithinBounds = period > lbConfig.MinPeriod && period < lbConfig.MaxPeriod;
//Console.WriteLine($"step={i} period={period:f4}");
if (!periodWithinBounds)
{
continue;
}
// lay down score for sample length
for (int j = 0; j < dctLength; j++)
{
if (scores[i + j] < intensity)
{
scores[i + j] = intensity;
}
}
if (maximumIntensity < intensity)
{
maximumIntensity = intensity;
}
}
// add abbreviatedSpeciesName into event
if (maximumIntensity >= intensityThreshold)
{
ae.Name = "L.b";
ae.Score_MaxInEvent = maximumIntensity;
confirmedEvents.Add(ae);
}
}
//######################################################################
// calculate the cosine similarity scores
var scorePlot = new Plot(lbConfig.SpeciesName, scores, intensityThreshold);
//DEBUG IMAGE this recognizer only. MUST set false for deployment.
Image debugImage = null;
if (drawDebugImage)
{
// display a variety of debug score arrays
//DataTools.Normalise(scores, eventDecibelThreshold, out normalisedScores, out normalisedThreshold);
//var debugPlot = new Plot("Score", normalisedScores, normalisedThreshold);
//DataTools.Normalise(upperArray, eventDecibelThreshold, out normalisedScores, out normalisedThreshold);
//var upperPlot = new Plot("Upper", normalisedScores, normalisedThreshold);
//DataTools.Normalise(lowerArray, eventDecibelThreshold, out normalisedScores, out normalisedThreshold);
//var lowerPlot = new Plot("Lower", normalisedScores, normalisedThreshold);
DataTools.Normalise(amplitudeScores, decibelThreshold, out var normalisedScores, out var normalisedThreshold);
var sumDiffPlot = new Plot("SumMinusDifference", normalisedScores, normalisedThreshold);
DataTools.Normalise(differenceScores, 3.0, out normalisedScores, out normalisedThreshold);
var differencePlot = new Plot("Difference", normalisedScores, normalisedThreshold);
var debugPlots = new List <Plot> {
scorePlot, sumDiffPlot, differencePlot
};
// other debug plots
//var debugPlots = new List<Plot> { scorePlot, upperPlot, lowerPlot, sumDiffPlot, differencePlot };
debugImage = DisplayDebugImage(sonogram, confirmedEvents, debugPlots, hits);
}
// return new sonogram because it makes for more easy interpretation of the image
var returnSonoConfig = new SonogramConfig
{
SourceFName = recording.BaseName,
WindowSize = 512,
WindowOverlap = 0,
// the default window is HAMMING
//WindowFunction = WindowFunctions.HANNING.ToString(),
//WindowFunction = WindowFunctions.NONE.ToString(),
// if do not use noise reduction can get a more sensitive recogniser.
//NoiseReductionType = NoiseReductionType.NONE,
NoiseReductionType = SNR.KeyToNoiseReductionType("STANDARD"),
};
BaseSonogram returnSonogram = new SpectrogramStandard(returnSonoConfig, recording.WavReader);
return(Tuple.Create(returnSonogram, hits, scores, confirmedEvents, debugImage));
} //Analysis()
} // end method ConvertODScores2Events()
/*
* public static double PeakEntropy(double[] array)
* {
* bool[] peaks = DataTools.GetPeaks(array);
* int peakCount = DataTools.CountTrues(peaks);
* //set up histogram of peak energies
* double[] histogram = new double[peakCount];
* int count = 0;
* for (int k = 0; k < array.Length; k++)
* {
* if (peaks[k])
* {
* histogram[count] = array[k];
* count++;
* }
* }
* histogram = DataTools.NormaliseMatrixValues(histogram);
* histogram = DataTools.Normalise2Probabilites(histogram);
* double normFactor = Math.Log(histogram.Length) / DataTools.ln2; //normalize for length of the array
* double entropy = DataTools.Entropy(histogram) / normFactor;
* return entropy;
* }
*
*/
/// <summary>
/// returns the periodicity in an array of values.
/// </summary>
public static double[] PeriodicityAnalysis(double[] array)
{
//DataTools.writeBarGraph(array);
var A = AutoAndCrossCorrelation.MyCrossCorrelation(array, array); // do 2/3rds of maximum possible lag
int dctLength = A.Length;
A = DataTools.SubtractMean(A);
//DataTools.writeBarGraph(A);
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
double[] dct = MFCCStuff.DCT(A, cosines);
for (int i = 0; i < dctLength; i++)
{
dct[i] = Math.Abs(dct[i]); //convert to absolute values
}
//DataTools.writeBarGraph(dct);
for (int i = 0; i < 3; i++)
{
dct[i] = 0.0; //remove low freq oscillations from consideration
}
dct = DataTools.normalise2UnitLength(dct);
var peaks = DataTools.GetPeaks(dct);
// remove non-peak values and low values
for (int i = 0; i < dctLength; i++)
{
if (!peaks[i] || dct[i] < 0.2)
{
dct[i] = 0.0;
}
}
DataTools.writeBarGraph(dct);
//get periodicity of highest three values
int peakCount = 3;
var period = new double[peakCount];
var maxIndex = new double[peakCount];
for (int i = 0; i < peakCount; i++)
{
int indexOfMaxValue = DataTools.GetMaxIndex(dct);
maxIndex[i] = indexOfMaxValue;
//double oscilFreq = indexOfMaxValue / dctDuration * 0.5; //Times 0.5 because index = Pi and not 2Pi
if ((double)indexOfMaxValue == 0)
{
period[i] = 0.0;
}
else
{
period[i] = dctLength / (double)indexOfMaxValue * 2;
}
dct[indexOfMaxValue] = 0.0; // remove value for next iteration
}
LoggedConsole.WriteLine("Max indices = {0:f0}, {1:f0}, {2:f0}.", maxIndex[0], maxIndex[1], maxIndex[2]);
return(period);
}
/*
*
* /// <summary>
* /// Detects oscillations in a given freq bin.
* /// there are several important parameters for tuning.
* /// a) DCTLength: Good values are 0.25 to 0.50 sec. Do not want too long because DCT requires stationarity.
* /// Do not want too short because too small a range of oscillations
* /// b) DCTindex: Sets lower bound for oscillations of interest. Index refers to array of coeff returned by DCT.
* /// Array has same length as the length of the DCT. Low freq oscillations occur more often by chance. Want to exclude them.
* /// c) MinAmplitude: minimum acceptable value of a DCT coefficient if hit is to be accepted.
* /// The algorithm is sensitive to this value. A lower value results in more oscillation hits being returned.
* /// </summary>
* /// <param name="minBin">min freq bin of search band</param>
* /// <param name="maxBin">max freq bin of search band</param>
* /// <param name="dctLength">number of values</param>
* /// <param name="DCTindex">Sets lower bound for oscillations of interest.</param>
* /// <param name="minAmplitude">threshold - do not accept a DCT value if its amplitude is less than this threshold</param>
* public static Double[,] DetectOscillations(SpectrogramStandard sonogram, int minHz, int maxHz,
* double dctDuration, int minOscilFreq, int maxOscilFreq, double minAmplitude)
* {
* int minBin = (int)(minHz / sonogram.FBinWidth);
* int maxBin = (int)(maxHz / sonogram.FBinWidth);
*
* int dctLength = (int)Math.Round(sonogram.FramesPerSecond * dctDuration);
* int minIndex = (int)(minOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
* int maxIndex = (int)(maxOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
* if (maxIndex > dctLength) maxIndex = dctLength; //safety check in case of future changes to code.
*
* int rows = sonogram.Data.GetLength(0);
* int cols = sonogram.Data.GetLength(1);
* Double[,] hits = new Double[rows, cols];
* Double[,] matrix = sonogram.Data;
* //matrix = ImageTools.WienerFilter(sonogram.Data, 3);// DO NOT USE - SMUDGES EVERYTHING
*
*
* double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
* //following two lines write matrix of cos values for checking.
* //string fPath = @"C:\SensorNetworks\Sonograms\cosines.txt";
* //FileTools.WriteMatrix2File_Formatted(cosines, fPath, "F3");
*
* //following two lines write bmp image of cos values for checking.
* //string fPath = @"C:\SensorNetworks\Output\cosines.bmp";
* //ImageTools.DrawMatrix(cosines, fPath);
*
*
*
* // traverse columns - skip DC column
*
*
* for (int c = minBin; c <= maxBin; c++) {
* for (int r = 0; r < rows - dctLength; r++)
* {
* var array = new double[dctLength];
* //accumulate J columns of values
* for (int i = 0; i < dctLength; i++)
* for (int j = 0; j < 5; j++) array[i] += matrix[r + i, c + j];
*
* array = DataTools.SubtractMean(array);
* // DataTools.writeBarGraph(array);
*
* double[] dct = MFCCStuff.DCT(array, cosines);
* for (int i = 0; i < dctLength; i++) dct[i] = Math.Abs(dct[i]);//convert to absolute values
* dct[0] = 0.0; dct[1] = 0.0; dct[2] = 0.0; dct[3] = 0.0; dct[4] = 0.0;//remove low freq oscillations from consideration
* dct = DataTools.normalise2UnitLength(dct);
* //dct = DataTools.NormaliseMatrixValues(dct); //another option to NormaliseMatrixValues
* int indexOfMaxValue = DataTools.GetMaxIndex(dct);
* double oscilFreq = indexOfMaxValue / dctDuration * 0.5; //Times 0.5 because index = Pi and not 2Pi
*
* //DataTools.MinMax(dct, out min, out max);
* // DataTools.writeBarGraph(dct);
*
* //mark DCT location with oscillation freq, only if oscillation freq is in correct range and amplitude
* if ((indexOfMaxValue >= minIndex) && (indexOfMaxValue <= maxIndex) && (dct[indexOfMaxValue] > minAmplitude))
* {
* for (int i = 0; i < dctLength; i++) hits[r + i, c] = oscilFreq;
* }
* r += 5; //skip rows
* }
* c++; //do alternate columns
* }
* return hits;
* }
*/
public static double[] DetectOscillationsInScoreArray(double[] scoreArray, double dctDuration, double timeScale, double dctThreshold,
bool normaliseDCT, int minOscilFreq, int maxOscilFreq)
{
int dctLength = (int)Math.Round(timeScale * dctDuration);
int minIndex = (int)(minOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int maxIndex = (int)(maxOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
if (maxIndex > dctLength)
{
maxIndex = dctLength; //safety check in case of future changes to code.
}
int length = scoreArray.Length;
double[] hits = new double[length];
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
//following two lines write matrix of cos values for checking.
//string fPath = @"C:\SensorNetworks\Sonograms\cosines.txt";
//FileTools.WriteMatrix2File_Formatted(cosines, fPath, "F3");
//following two lines write bmp image of cos values for checking.
//string fPath = @"C:\SensorNetworks\Output\cosines.bmp";
//ImageTools.DrawMatrix(cosines, fPath);
for (int r = 0; r < length - dctLength; r++)
{
var array = new double[dctLength];
//transfer values
for (int i = 0; i < dctLength; i++)
{
array[i] = scoreArray[r + i];
}
array = DataTools.SubtractMean(array);
// DataTools.writeBarGraph(array);
double[] dct = MFCCStuff.DCT(array, cosines);
for (int i = 0; i < dctLength; i++)
{
dct[i] = Math.Abs(dct[i]); //convert to absolute values
}
for (int i = 0; i < 5; i++)
{
dct[i] = 0.0; //remove low freq oscillations from consideration
}
if (normaliseDCT)
{
dct = DataTools.normalise2UnitLength(dct);
}
int indexOfMaxValue = DataTools.GetMaxIndex(dct);
double oscilFreq = indexOfMaxValue / dctDuration * 0.5; //Times 0.5 because index = Pi and not 2Pi
// DataTools.writeBarGraph(dct);
//LoggedConsole.WriteLine("oscilFreq = " + oscilFreq);
//mark DCT location with oscillation freq, only if oscillation freq is in correct range and amplitude
if (indexOfMaxValue >= minIndex && indexOfMaxValue <= maxIndex && dct[indexOfMaxValue] > dctThreshold)
{
hits[r] = dct[indexOfMaxValue];
hits[r + 1] = dct[indexOfMaxValue]; // because skipping rows.
//for (int i = 0; i < dctLength; i++) if (hits[r + i] < dct[indexOfMaxValue]) hits[r + i] = dct[indexOfMaxValue];
}
r += 1; //skip rows
}
return(hits);
}
/*
* /// <summary>
* /// FINDS OSCILLATIONS IN A SONOGRAM
* /// SAME METHOD AS ABOVE BUT .....
* /// 1) WITHOUT CALCULATING THE COMPUTATION TIME
* /// 2) WITHOUT DOING SEGMENTATION
* /// </summary>
* /// <param name="sonogram">sonogram derived from the recording</param>
* /// <param name="minHz">min bound freq band to search</param>
* /// <param name="maxHz">max bound freq band to search</param>
* /// <param name="dctDuration">duration of DCT in seconds</param>
* /// <param name="dctThreshold">minimum amplitude of DCT </param>
* /// <param name="minOscilFreq">ignore oscillation frequencies below this threshold</param>
* /// <param name="maxOscilFreq">ignore oscillation frequencies greater than this </param>
* /// <param name="scoreThreshold">used for FP/FN</param>
* /// <param name="minDuration">ignore hits whose duration is shorter than this</param>
* /// <param name="maxDuration">ignore hits whose duration is longer than this</param>
* /// <param name="scores">return an array of scores over the entire recording</param>
* /// <param name="events">return a list of acoustic events</param>
* /// <param name="hits">a matrix to be superimposed over the final sonogram which shows where the DCT coefficients exceeded the threshold</param>
* public static void Execute(SpectrogramStandard sonogram, int minHz, int maxHz,
* double dctDuration, double dctThreshold, bool normaliseDCT, double minOscilFreq, double maxOscilFreq,
* double scoreThreshold, double minDuration, double maxDuration,
* out double[] scores, out List<AcousticEvent> events, out Double[,] hits, out double[] oscFreq)
* {
* //convert the entire recording to an acoustic event - this is the legacy of previous experimentation!!!!!!!!!
* List<AcousticEvent> segmentEvents = new List<AcousticEvent>();
* var ae = new AcousticEvent(0.0, sonogram.Duration.TotalSeconds, minHz, maxHz);
* ae.SetTimeAndFreqScales(sonogram.FramesPerSecond, sonogram.FBinWidth);
* segmentEvents.Add(ae);
*
* //DETECT OSCILLATIONS
* hits = DetectOscillationsInSonogram(sonogram, minHz, maxHz, dctDuration, dctThreshold, normaliseDCT, minOscilFreq, maxOscilFreq, segmentEvents);
* hits = RemoveIsolatedOscillations(hits);
*
* //EXTRACT SCORES AND ACOUSTIC EVENTS
* scores = GetOscillationScores(hits, minHz, maxHz, sonogram.FBinWidth);//scores = fraction of BW bins in each row that have an oscilation hit.
* scores = DataTools.filterMovingAverage(scores, 3);
* oscFreq = GetOscillationFrequency(hits, minHz, maxHz, sonogram.FBinWidth);
* events = ConvertODScores2Events(scores, oscFreq, minHz, maxHz, sonogram.FramesPerSecond, sonogram.FBinWidth, sonogram.Configuration.FreqBinCount, scoreThreshold,
* minDuration, maxDuration, sonogram.Configuration.SourceFName);
* }
*
*/
/// <summary>
/// Detects oscillations in a given freq bin.
/// there are several important parameters for tuning.
/// a) dctDuration: Good values are 0.25 to 0.50 sec. Do not want too long because DCT requires stationarity.
/// Do not want too short because too small a range of oscillations
/// b) dctThreshold: minimum acceptable value of a DCT coefficient if hit is to be accepted.
/// The algorithm is sensitive to this value. A lower value results in more oscillation hits being returned.
/// c) Min and Max Oscillaitons: Sets lower & upper bound for oscillations of interest.
/// Array has same length as the length of the DCT. Low freq oscillations occur more often by chance. Want to exclude them.
/// </summary>
/// <param name="minHz">min freq bin of search band.</param>
/// <param name="maxHz">max freq bin of search band.</param>
public static double[,] DetectOscillationsInSonogram(SpectrogramStandard sonogram, int minHz, int maxHz, double dctDuration, double dctThreshold,
bool normaliseDCT, double minOscilFreq, double maxOscilFreq, List <AcousticEvent> events)
{
if (events == null)
{
return(null);
}
int minBin = (int)(minHz / sonogram.FBinWidth);
int maxBin = (int)(maxHz / sonogram.FBinWidth);
int dctLength = (int)Math.Round(sonogram.FramesPerSecond * dctDuration);
int minIndex = (int)(minOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int maxIndex = (int)(maxOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
if (maxIndex > dctLength)
{
maxIndex = dctLength; //safety check in case of future changes to code.
}
int rows = sonogram.Data.GetLength(0);
int cols = sonogram.Data.GetLength(1);
double[,] hits = new double[rows, cols];
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
//following two lines write matrix of cos values for checking.
//string fPath = @"C:\SensorNetworks\Sonograms\cosines.txt";
//FileTools.WriteMatrix2File_Formatted(cosines, fPath, "F3");
//following two lines write bmp image of cos values for checking.
string fPath = @"C:\SensorNetworks\Output\cosines.bmp";
ImageTools.DrawMatrix(cosines, fPath, true);
foreach (AcousticEvent av in events)
{
int startRow = (int)Math.Round(av.TimeStart * sonogram.FramesPerSecond);
int endRow = (int)Math.Round(av.TimeEnd * sonogram.FramesPerSecond);
if (endRow >= sonogram.FrameCount)
{
endRow = sonogram.FrameCount - 1;
}
endRow -= dctLength;
if (endRow <= startRow)
{
endRow = startRow + 1; //want minimum of one row
}
// traverse columns
for (int c = minBin; c <= maxBin; c++)
{
for (int r = startRow; r < endRow; r++)
{
var array = new double[dctLength];
//accumulate J columns of values
int N = 5; //average five rows
for (int i = 0; i < dctLength; i++)
{
for (int j = 0; j < N; j++)
{
array[i] += sonogram.Data[r + i, c + j];
}
array[i] /= N;
}
array = DataTools.SubtractMean(array);
// DataTools.writeBarGraph(array);
//double entropy = PeakEntropy(array);
//if (entropy < 0.85)
//{
// r += 6; //skip rows
// continue;
//}
int lowFreqBuffer = 5;
double[] dct = MFCCStuff.DCT(array, cosines);
for (int i = 0; i < dctLength; i++)
{
dct[i] = Math.Abs(dct[i]); //convert to absolute values
}
for (int i = 0; i < lowFreqBuffer; i++)
{
dct[i] = 0.0; //remove low freq oscillations from consideration
}
if (normaliseDCT)
{
dct = DataTools.normalise2UnitLength(dct);
}
int indexOfMaxValue = DataTools.GetMaxIndex(dct);
double oscilFreq = indexOfMaxValue / dctDuration * 0.5; //Times 0.5 because index = Pi and not 2Pi
//DataTools.writeBarGraph(dct);
//LoggedConsole.WriteLine("oscilFreq ={0:f2} (max index={1}) Amp={2:f2}", oscilFreq, indexOfMaxValue, dct[indexOfMaxValue]);
//calculate specificity i.e. what other oscillations are present.
//double offMaxAmplitude = 0.0;
//for (int i = lowFreqBuffer; i < dctLength; i++) offMaxAmplitude += dct[i];
//offMaxAmplitude -= (dct[indexOfMaxValue-1] + dct[indexOfMaxValue] + dct[indexOfMaxValue+1]);
//offMaxAmplitude /= (dctLength - lowFreqBuffer - 3); //get average
//double specificity = 2 * (0.5 - (offMaxAmplitude / dct[indexOfMaxValue]));
////LoggedConsole.WriteLine("avOffAmp={0:f2} specificity ={1:f2}", offMaxAmplitude, specificity);
//double threshold = dctThreshold + dctThreshold;
//mark DCT location with oscillation freq, only if oscillation freq is in correct range and amplitude
if (indexOfMaxValue >= minIndex && indexOfMaxValue <= maxIndex && dct[indexOfMaxValue] > dctThreshold)
//if ((indexOfMaxValue >= minIndex) && (indexOfMaxValue <= maxIndex) && ((dct[indexOfMaxValue] * specificity) > threshold))
{
for (int i = 0; i < dctLength; i++)
{
hits[r + i, c] = oscilFreq;
}
for (int i = 0; i < dctLength; i++)
{
hits[r + i, c + 1] = oscilFreq; //write alternate column - MUST DO THIS BECAUSE doing alternate columns
}
}
r += 6; //skip rows i.e. frames of the sonogram.
}
c++; //do alternate columns
}
} //foreach (AcousticEvent av in events)
return(hits);
}
public static double[] DetectOscillations(double[] ipArray, double framesPerSecond, double dctDuration, double minOscilFreq, double maxOscilFreq, double dctThreshold)
{
int dctLength = (int)Math.Round(framesPerSecond * dctDuration);
int minIndex = (int)(minOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int maxIndex = (int)(maxOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
//double midOscilFreq = minOscilFreq + ((maxOscilFreq - minOscilFreq) / 2);
if (maxIndex > dctLength)
{
return(null); //safety check
}
int length = ipArray.Length;
var dctScores = new double[length];
//var hits = new double[length];
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
//following two lines write bmp image of cosine matrix values for checking.
//string bmpPath = @"C:\SensorNetworks\Output\cosines.png";
//ImageTools.DrawMatrix(cosines, bmpPath, true);
for (int r = 1; r < length - dctLength; r++)
{
// only stop if current location is a peak
if (ipArray[r] < ipArray[r - 1] || ipArray[r] < ipArray[r + 1])
{
continue;
}
// extract array and ready for DCT
//for (int i = 0; i < dctLength; i++) dctArray[i] = ipArray[r + i];
var dctArray = DataTools.Subarray(ipArray, r, dctLength);
dctArray = DataTools.SubtractMean(dctArray);
//dctArray = DataTools.Vector2Zscores(dctArray);
double[] dctCoeff = MFCCStuff.DCT(dctArray, cosines);
// convert to absolute values because not interested in negative values due to phase.
for (int i = 0; i < dctLength; i++)
{
dctCoeff[i] = Math.Abs(dctCoeff[i]);
}
// remove low freq oscillations from consideration
int thresholdIndex = minIndex / 4;
for (int i = 0; i < thresholdIndex; i++)
{
dctCoeff[i] = 0.0;
}
dctCoeff = DataTools.normalise2UnitLength(dctCoeff);
//dct = DataTools.NormaliseMatrixValues(dct); //another option to NormaliseMatrixValues
int indexOfMaxValue = DataTools.GetMaxIndex(dctCoeff);
//double oscilFreq = indexOfMaxValue / dctDuration * 0.5; //Times 0.5 because index = Pi and not 2Pi
// #### Tried this option for scoring oscillation hits but did not work well.
// #### Requires very fine tuning of thresholds
//dctCoeff = DataTools.Normalise2Probabilites(dctCoeff);
//// sum area under curve where looking for oscillations
//double sum = 0.0;
//for (int i = minIndex; i <= maxIndex; i++)
// sum += dctCoeff[i];
//if (sum > dctThreshold)
//{
// for (int i = 0; i < dctLength; i++) hits[r + i, c] = midOscilFreq;
//}
// DEBUGGING
// DataTools.MinMax(dctCoeff, out min, out max);
//DataTools.writeBarGraph(dctArray);
//DataTools.writeBarGraph(dctCoeff);
//mark DCT location with oscillation freq, only if oscillation freq is in correct range and amplitude
if (indexOfMaxValue >= minIndex && indexOfMaxValue <= maxIndex && dctCoeff[indexOfMaxValue] > dctThreshold)
{
//for (int i = 0; i < dctLength; i++) dctScores[r + i] = midOscilFreq;
for (int i = 0; i < dctLength; i++)
{
if (dctScores[r + i] < dctCoeff[indexOfMaxValue])
{
dctScores[r + i] = dctCoeff[indexOfMaxValue];
}
}
}
}
//return hits; //dctArray
return(dctScores);
}
/// <summary>
/// Detects oscillations in a given freq bin.
/// there are several important parameters for tuning.
/// a) DCTLength: Good values are 0.25 to 0.50 sec. Do not want too long because DCT requires stationarity.
/// Do not want too short because too small a range of oscillations
/// b) DCTindex: Sets lower bound for oscillations of interest. Index refers to array of coeff returned by DCT.
/// Array has same length as the length of the DCT. Low freq oscillations occur more often by chance. Want to exclude them.
/// c) MinAmplitude: minimum acceptable value of a DCT coefficient if hit is to be accepted.
/// The algorithm is sensitive to this value. A lower value results in more oscillation hits being returned.
/// </summary>
/// <param name="sonogram"></param>
/// <param name="minHz">min freq bin of search band</param>
/// <param name="maxHz">max freq bin of search band</param>
/// <param name="dctDuration">number of values</param>
/// <param name="maxOscilFreq"></param>
/// <param name="dctThreshold">threshold - do not accept a DCT coefficient if its value is less than this threshold</param>
/// <param name="minOscilFreq"></param>
/// <returns></returns>
public static double[,] DetectOscillations(SpectrogramStandard sonogram, int minHz, int maxHz,
double dctDuration, int minOscilFreq, int maxOscilFreq, double dctThreshold)
{
int minBin = (int)(minHz / sonogram.FBinWidth);
int maxBin = (int)(maxHz / sonogram.FBinWidth);
int dctLength = (int)Math.Round(sonogram.FramesPerSecond * dctDuration);
int minIndex = (int)(minOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int maxIndex = (int)(maxOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int midOscilFreq = minOscilFreq + ((maxOscilFreq - minOscilFreq) / 2);
if (maxIndex > dctLength)
{
return(null); //safety check
}
int rows = sonogram.Data.GetLength(0);
int cols = sonogram.Data.GetLength(1);
double[,] hits = new double[rows, cols];
double[,] matrix = sonogram.Data;
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
//following two lines write matrix of cos values for checking.
//string txtPath = @"C:\SensorNetworks\Output\cosines.txt";
//FileTools.WriteMatrix2File_Formatted(cosines, txtPath, "F3");
//following two lines write bmp image of cos values for checking.
//string bmpPath = @"C:\SensorNetworks\Output\cosines.png";
//ImageTools.DrawMatrix(cosines, bmpPath, true);
for (int c = minBin; c <= maxBin; c++) //traverse columns - skip DC column
{
var dctArray = new double[dctLength];
for (int r = 0; r < rows - dctLength; r++)
{
// extract array and ready for DCT
for (int i = 0; i < dctLength; i++)
{
dctArray[i] = matrix[r + i, c];
}
dctArray = DataTools.SubtractMean(dctArray);
//dctArray = DataTools.Vector2Zscores(dctArray);
double[] dctCoeff = MFCCStuff.DCT(dctArray, cosines);
// convert to absolute values because not interested in negative values due to phase.
for (int i = 0; i < dctLength; i++)
{
dctCoeff[i] = Math.Abs(dctCoeff[i]);
}
// remove low freq oscillations from consideration
int thresholdIndex = minIndex / 4;
for (int i = 0; i < thresholdIndex; i++)
{
dctCoeff[i] = 0.0;
}
dctCoeff = DataTools.normalise2UnitLength(dctCoeff);
//dct = DataTools.NormaliseMatrixValues(dct); //another option to NormaliseMatrixValues
int indexOfMaxValue = DataTools.GetMaxIndex(dctCoeff);
//double oscilFreq = indexOfMaxValue / dctDuration * 0.5; //Times 0.5 because index = Pi and not 2Pi
// #### Tried this option for scoring oscillation hits but did not work well.
// #### Requires very fine tuning of thresholds
//dctCoeff = DataTools.Normalise2Probabilites(dctCoeff);
//// sum area under curve where looking for oscillations
//double sum = 0.0;
//for (int i = minIndex; i <= maxIndex; i++)
// sum += dctCoeff[i];
//if (sum > dctThreshold)
//{
// for (int i = 0; i < dctLength; i++) hits[r + i, c] = midOscilFreq;
//}
// DEBUGGING
// DataTools.MinMax(dctCoeff, out min, out max);
//DataTools.writeBarGraph(dctArray);
//DataTools.writeBarGraph(dctCoeff);
//mark DCT location with oscillation freq, only if oscillation freq is in correct range and amplitude
if (indexOfMaxValue >= minIndex && indexOfMaxValue <= maxIndex && dctCoeff[indexOfMaxValue] > dctThreshold)
{
for (int i = 0; i < dctLength; i++)
{
hits[r + i, c] = midOscilFreq;
}
}
r += 5; //skip rows
}
c++; //do alternate columns
}
return(hits);
}
/// <summary>
/// Detects oscillations in a given freq bin.
/// there are several important parameters for tuning.
/// a) DCTLength: Good values are 0.25 to 0.50 sec. Do not want too long because DCT requires stationarity.
/// Do not want too short because too small a range of oscillations
/// b) DCTindex: Sets lower bound for oscillations of interest. Index refers to array of coefficient returned by DCT.
/// Array has same length as the length of the DCT. Low freq oscillations occur more often by chance. Want to exclude them.
/// c) MinAmplitude: minimum acceptable value of a DCT coefficient if hit is to be accepted.
/// The algorithm is sensitive to this value. A lower value results in more oscillation hits being returned.
/// </summary>
/// <param name="sonogram">A spectrogram.</param>
/// <param name="minHz">min freq bin of search band.</param>
/// <param name="maxHz">max freq bin of search band.</param>
/// <param name="dctDuration">number of values.</param>
/// <param name="minOscilFreq">minimum oscillation freq.</param>
/// <param name="maxOscilFreq">maximum oscillation freq.</param>
/// <param name="dctThreshold">threshold - do not accept a DCT coefficient if its value is less than this threshold.</param>
public static double[,] DetectOscillations(SpectrogramStandard sonogram, int minHz, int maxHz, double dctDuration, int minOscilFreq, int maxOscilFreq, double dctThreshold)
{
int minBin = (int)(minHz / sonogram.FBinWidth);
int maxBin = (int)(maxHz / sonogram.FBinWidth);
int dctLength = (int)Math.Round(sonogram.FramesPerSecond * dctDuration);
int minIndex = (int)(minOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int maxIndex = (int)(maxOscilFreq * dctDuration * 2); //multiply by 2 because index = Pi and not 2Pi
int midOscilFreq = minOscilFreq + ((maxOscilFreq - minOscilFreq) / 2);
//safety check
if (maxIndex > dctLength)
{
return(null);
}
int rows = sonogram.Data.GetLength(0);
int cols = sonogram.Data.GetLength(1);
double[,] hits = new double[rows, cols];
double[,] matrix = sonogram.Data;
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength); //set up the cosine coefficients
//following two lines write matrix of cos values for checking.
//string txtPath = @"C:\SensorNetworks\Output\cosines.txt";
//FileTools.WriteMatrix2File_Formatted(cosines, txtPath, "F3");
//following two lines write bmp image of cos values for checking.
//string bmpPath = @"C:\SensorNetworks\Output\cosines.png";
//ImageTools.DrawMatrix(cosines, bmpPath, true);
//traverse columns - skip DC column
for (int c = minBin; c <= maxBin; c++)
{
var dctArray = new double[dctLength];
for (int r = 0; r < rows - dctLength; r++)
{
// extract array and ready for DCT
for (int i = 0; i < dctLength; i++)
{
dctArray[i] = matrix[r + i, c];
}
int lowerDctBound = minIndex / 4;
var dctCoeff = DoDct(dctArray, cosines, lowerDctBound);
int indexOfMaxValue = DataTools.GetMaxIndex(dctCoeff);
//mark DCT location with oscillation freq, only if oscillation freq is in correct range and amplitude
if (indexOfMaxValue >= minIndex && indexOfMaxValue <= maxIndex && dctCoeff[indexOfMaxValue] > dctThreshold)
{
for (int i = 0; i < dctLength; i++)
{
hits[r + i, c] = midOscilFreq;
}
}
r += 5; //skip rows
}
c++; //do alternate columns
}
return(hits);
}
/// <summary>
/// Do your analysis. This method is called once per segment (typically one-minute segments).
/// </summary>
public override RecognizerResults Recognize(AudioRecording recording, Config configuration, TimeSpan segmentStartOffset, Lazy <IndexCalculateResult[]> getSpectralIndexes, DirectoryInfo outputDirectory, int?imageWidth)
{
string speciesName = configuration[AnalysisKeys.SpeciesName] ?? "<no species>";
string abbreviatedSpeciesName = configuration[AnalysisKeys.AbbreviatedSpeciesName] ?? "<no.sp>";
const int frameSize = 256;
const double windowOverlap = 0.0;
double noiseReductionParameter = configuration.GetDoubleOrNull("SeverityOfNoiseRemoval") ?? 2.0;
int minHz = configuration.GetInt(AnalysisKeys.MinHz);
int maxHz = configuration.GetInt(AnalysisKeys.MaxHz);
// ignore oscillations below this threshold freq
int minOscilFreq = configuration.GetInt(AnalysisKeys.MinOscilFreq);
// ignore oscillations above this threshold freq
int maxOscilFreq = configuration.GetInt(AnalysisKeys.MaxOscilFreq);
// duration of DCT in seconds
//double dctDuration = (double)configuration[AnalysisKeys.DctDuration];
// minimum acceptable value of a DCT coefficient
double dctThreshold = configuration.GetDouble(AnalysisKeys.DctThreshold);
// min duration of event in seconds
double minDuration = configuration.GetDouble(AnalysisKeys.MinDuration);
// max duration of event in seconds
double maxDuration = configuration.GetDouble(AnalysisKeys.MaxDuration);
// min score for an acceptable event
double decibelThreshold = configuration.GetDouble(AnalysisKeys.DecibelThreshold);
// min score for an acceptable event
double eventThreshold = configuration.GetDouble(AnalysisKeys.EventThreshold);
if (recording.WavReader.SampleRate != 22050)
{
throw new InvalidOperationException("Requires a 22050Hz file");
}
// i: MAKE SONOGRAM
var sonoConfig = new SonogramConfig
{
SourceFName = recording.BaseName,
WindowSize = frameSize,
WindowOverlap = windowOverlap,
NoiseReductionType = NoiseReductionType.Standard,
NoiseReductionParameter = noiseReductionParameter,
};
var recordingDuration = recording.Duration;
int sr = recording.SampleRate;
double freqBinWidth = sr / (double)sonoConfig.WindowSize;
int minBin = (int)Math.Round(minHz / freqBinWidth) + 1;
int maxBin = (int)Math.Round(maxHz / freqBinWidth) + 1;
// duration of DCT in seconds - want it to be about 3X or 4X the expected maximum period
double framesPerSecond = freqBinWidth;
double minPeriod = 1 / (double)maxOscilFreq;
double maxPeriod = 1 / (double)minOscilFreq;
double dctDuration = 5 * maxPeriod;
// duration of DCT in frames
int dctLength = (int)Math.Round(framesPerSecond * dctDuration);
// set up the cosine coefficients
double[,] cosines = MFCCStuff.Cosines(dctLength, dctLength);
BaseSonogram sonogram = new SpectrogramStandard(sonoConfig, recording.WavReader);
int rowCount = sonogram.Data.GetLength(0);
double[] amplitudeArray = MatrixTools.GetRowAveragesOfSubmatrix(sonogram.Data, 0, minBin, rowCount - 1, maxBin);
// remove baseline from amplitude array
var highPassFilteredSignal = DspFilters.SubtractBaseline(amplitudeArray, 7);
// remove hi freq content from amplitude array
var lowPassFilteredSignal = DataTools.filterMovingAverageOdd(amplitudeArray, 11);
var dctScores = new double[highPassFilteredSignal.Length];
const int step = 2;
for (int i = dctLength; i < highPassFilteredSignal.Length - dctLength; i += step)
{
if (highPassFilteredSignal[i] < decibelThreshold)
{
continue;
}
double[] subArray = DataTools.Subarray(highPassFilteredSignal, i, dctLength);
// Look for oscillations in the highPassFilteredSignal
Oscillations2014.GetOscillationUsingDct(subArray, framesPerSecond, cosines, out var oscilFreq, out var period, out var intensity);
bool periodWithinBounds = period > minPeriod && period < maxPeriod;
if (!periodWithinBounds)
{
continue;
}
if (intensity < dctThreshold)
{
continue;
}
//lay down score for sample length
for (int j = 0; j < dctLength; j++)
{
if (dctScores[i + j] < intensity && lowPassFilteredSignal[i + j] > decibelThreshold)
{
dctScores[i + j] = intensity;
}
}
}
//iii: CONVERT decibel sum-diff SCORES TO ACOUSTIC EVENTS
var acousticEvents = AcousticEvent.ConvertScoreArray2Events(
dctScores,
minHz,
maxHz,
sonogram.FramesPerSecond,
freqBinWidth,
eventThreshold,
minDuration,
maxDuration,
segmentStartOffset);
// ######################################################################
acousticEvents.ForEach(ae =>
{
ae.SpeciesName = speciesName;
ae.SegmentDurationSeconds = recordingDuration.TotalSeconds;
ae.SegmentStartSeconds = segmentStartOffset.TotalSeconds;
ae.Name = abbreviatedSpeciesName;
});
var plot = new Plot(this.DisplayName, dctScores, eventThreshold);
var plots = new List <Plot> {
plot
};
// DEBUG IMAGE this recognizer only. MUST set false for deployment.
bool displayDebugImage = MainEntry.InDEBUG;
if (displayDebugImage)
{
// display a variety of debug score arrays
DataTools.Normalise(amplitudeArray, decibelThreshold, out var normalisedScores, out var normalisedThreshold);
var ampltdPlot = new Plot("amplitude", normalisedScores, normalisedThreshold);
DataTools.Normalise(highPassFilteredSignal, decibelThreshold, out normalisedScores, out normalisedThreshold);
var demeanedPlot = new Plot("Hi Pass", normalisedScores, normalisedThreshold);
DataTools.Normalise(lowPassFilteredSignal, decibelThreshold, out normalisedScores, out normalisedThreshold);
var lowPassPlot = new Plot("Low Pass", normalisedScores, normalisedThreshold);
var debugPlots = new List <Plot> {
ampltdPlot, lowPassPlot, demeanedPlot, plot
};
Image debugImage = SpectrogramTools.GetSonogramPlusCharts(sonogram, acousticEvents, debugPlots, null);
var debugPath = outputDirectory.Combine(FilenameHelpers.AnalysisResultName(Path.GetFileNameWithoutExtension(recording.BaseName), this.Identifier, "png", "DebugSpectrogram"));
debugImage.Save(debugPath.FullName);
}
return(new RecognizerResults()
{
Sonogram = sonogram,
Hits = null,
Plots = plots,
Events = acousticEvents,
});
}