/// <summary>
/// Returns prediction for a sample at position
/// </summary>
public static float CalcBurgPred(
AudioData audioData,
int position)
{
var historyLengthSamples =
audioData.AudioProcessingSettings.HistoryLengthSamples;
// use output audio as an input because it already contains
// fixed samples before sample at position
var audioShort = new double[historyLengthSamples + 1];
for (var index = 0; index < historyLengthSamples + 1; index++)
{
audioShort[index] = audioData.GetOutputSample(
position - historyLengthSamples + index);
}
var fba = new FastBurgAlgorithm64(audioShort);
fba.Train(historyLengthSamples,
audioData.AudioProcessingSettings.CoefficientsNumber * 2,
historyLengthSamples);
return((float)fba.GetForwardPrediction());
}
private static float CalcBurgPredFromInput(
AudioData audioData,
int position)
{
var historyLengthSamples =
audioData.AudioProcessingSettings.HistoryLengthSamples;
var audioShort = new double[historyLengthSamples + 1];
for (var index = 0; index < historyLengthSamples + 1; index++)
{
audioShort[index] = audioData.GetInputSample(position -
historyLengthSamples + index);
}
var fba = new FastBurgAlgorithm64(audioShort);
fba.Train(historyLengthSamples,
audioData.AudioProcessingSettings.CoefficientsNumber * 2,
historyLengthSamples);
return((float)fba.GetForwardPrediction());
}
/// <summary>
/// Calculates prediction errors for a channel using CPU (Parallel.For)
/// </summary>
private static void CalculateBurgPredictionErrCpu(
AudioData audioData,
IProgress <double> progress)
{
var historyLengthSamples =
audioData.AudioProcessingSettings.HistoryLengthSamples;
var forwardPredictions = new float[audioData.LengthSamples()];
var backwardPredictions = new float[audioData.LengthSamples()];
var inputaudio = new double[audioData.LengthSamples()];
for (var index = 0; index < audioData.LengthSamples(); index++)
{
inputaudio[index] = audioData.GetInputSample(index);
}
// we will use steps to report progress
var step = audioData.LengthSamples() / 100;
for (var index = historyLengthSamples + 1;
index <= audioData.LengthSamples();
index += step)
{
progress.Report((double)100 * index / audioData.LengthSamples());
var endPosition = index + step;
if (endPosition > audioData.LengthSamples())
{
endPosition = audioData.LengthSamples();
}
Parallel.For(index, endPosition,
indexParallelFor =>
{
var fba = new FastBurgAlgorithm64(inputaudio);
fba.Train(indexParallelFor,
audioData.AudioProcessingSettings.CoefficientsNumber,
audioData.AudioProcessingSettings.HistoryLengthSamples);
forwardPredictions[indexParallelFor] = (float)fba.GetForwardPrediction();
backwardPredictions[indexParallelFor -
audioData.AudioProcessingSettings.HistoryLengthSamples -
1] =
(float)fba.GetBackwardPrediction();
}
);
}
progress.Report(0);
// for first samples forward predictions were not calculated
// we use backward predictions only
for (var index = 0; index < historyLengthSamples; index++)
{
audioData.SetPredictionErr(index,
audioData.GetInputSample(index) -
backwardPredictions[index]);
}
// finds prediction error based on forward and backward predictions
for (var index = historyLengthSamples;
index < audioData.LengthSamples() - historyLengthSamples;
index++)
{
audioData.SetPredictionErr(index,
audioData.GetInputSample(index) -
(forwardPredictions[index] +
backwardPredictions[index]) / 2);
}
// for last samples backward predictions were not calculated
// we use forward predictions only
for (var index = historyLengthSamples;
index < audioData.LengthSamples() - historyLengthSamples;
index++)
{
audioData.SetPredictionErr(index,
audioData.GetInputSample(index) -
forwardPredictions[index]);
}
}
