// This method makes an approximation, namely that the pitch mark interval is roughly constant.
// Usually, this will give a duration accurate to a few per cent (sufficient!) relative to the desired duration.
public WAVSound ChangeDuration(WAVSound sound, List <double> pitchMarkTimeList, double relativeDuration)
{
/* List<double> pitchPeriodList = new List<double>();
* for (int ii = 1; ii < pitchMarkTimeList.Count; ii++)
* {
* double pitchPeriod = pitchMarkTimeList[ii] - pitchMarkTimeList[ii - 1];
* pitchPeriodList.Add(pitchPeriod);
* }
* double averagePitchPeriod = pitchPeriodList.Average(); */
List <short> newSamples = new List <short>();
int firstPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTimeList[0]);
for (int ii = 0; ii < firstPitchMarkIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
if (relativeDuration <= 1)
{
int removalStepInterval = (int)Math.Round(1 / (1 - relativeDuration));
int pitchIndex = 1;
while (pitchIndex < pitchMarkTimeList.Count)
{
if ((pitchIndex % removalStepInterval) == 0)
{
// Nothing to do here: Simply avoid adding these samples
}
else
{
int previousPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTimeList[pitchIndex - 1]);
int currentPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTimeList[pitchIndex]);
for (int ii = previousPitchMarkIndex; ii < currentPitchMarkIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
}
pitchIndex++;
}
}
else if (relativeDuration > 1)
{
int additionStepInterval = (int)Math.Round(1 / (relativeDuration - 1));
int pitchIndex = 1;
while (pitchIndex < pitchMarkTimeList.Count)
{
if ((pitchIndex % additionStepInterval) == 0)
{
// Insert the samples for this pitch period twice:
int previousPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTimeList[pitchIndex - 1]);
int currentPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTimeList[pitchIndex]);
for (int ii = previousPitchMarkIndex; ii < currentPitchMarkIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
for (int ii = previousPitchMarkIndex; ii < currentPitchMarkIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
}
else
{
int previousPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTimeList[pitchIndex - 1]);
int currentPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTimeList[pitchIndex]);
for (int ii = previousPitchMarkIndex; ii < currentPitchMarkIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
}
pitchIndex++;
}
}
// Finally, build the sound from the new samples:
WAVSound newSound = new WAVSound(sound.Name, sound.SampleRate, sound.NumberOfChannels, sound.BitsPerSample);
newSound.GenerateFromSamples(new List <List <short> >()
{
newSamples, newSamples
});
return(newSound);
}
// Note: it is assumed that both channels (left and right) are equal.
public WAVSound ChangePitch(WAVSound sound, List <double> pitchMarkTimeList, double relativeStartPitch, double relativeEndPitch)
{
// First find the pitch mark indices in the original sound:
List <int> originalPitchMarkIndexList = new List <int>();
foreach (double pitchMarkTime in pitchMarkTimeList)
{
int originalPitchMarkIndex = sound.GetSampleIndexAtTime(pitchMarkTime);
originalPitchMarkIndexList.Add(originalPitchMarkIndex);
}
// Next, compute the index spacings of the pitch marks in the modified sound:
double originalSoundDuration = sound.GetDuration();
List <int> modifiedPitchMarkIndexSpacingList = new List <int>();
modifiedPitchMarkTimeList = new List <double>();
double firstModifiedPitchMarkTime = pitchMarkTimeList[0]; // First pitch mark unchanged
modifiedPitchMarkTimeList.Add(firstModifiedPitchMarkTime);
for (int ii = 1; ii < originalPitchMarkIndexList.Count; ii++)
{
int originalPitchMarkSpacing = originalPitchMarkIndexList[ii] - originalPitchMarkIndexList[ii - 1];
double relativePitch = relativeStartPitch + (pitchMarkTimeList[ii] / originalSoundDuration) * (relativeEndPitch - relativeStartPitch);
int modifiedPitchMarkIndexSpacing = (int)Math.Round(originalPitchMarkSpacing / relativePitch);
modifiedPitchMarkIndexSpacingList.Add(modifiedPitchMarkIndexSpacing);
double modifiedPitchMarkTime = modifiedPitchMarkTimeList.Last() + (double)modifiedPitchMarkIndexSpacing / (double)sound.SampleRate;
modifiedPitchMarkTimeList.Add(modifiedPitchMarkTime);
}
// Now build the sound, keeping the original sound data over a fraction (topFraction) of the pitch periods
// and interpolating between pitch periods:
List <short> newSamples = new List <short>();
// Special treatment of the first pitch period:
int firstPitchMarkIndex = originalPitchMarkIndexList[0]; // Position of the first pitch mark in the original sound
int firstModifiedPitchMarkIndexSpacing = modifiedPitchMarkIndexSpacingList[0]; // Spacing between the first and second pitch mark in the modified sound
int firstTopEndIndex = firstPitchMarkIndex + (int)Math.Round(topFraction * firstModifiedPitchMarkIndexSpacing);
for (int ii = 0; ii < firstTopEndIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
for (int iPitchMark = 1; iPitchMark < originalPitchMarkIndexList.Count; iPitchMark++)
{
// First add samples for the transition from the previous pitch period to the current one:
int modifiedPitchMarkIndexSpacing = modifiedPitchMarkIndexSpacingList[iPitchMark - 1]; // -1 since there are n-1 spacings for n pitch marks
int transitionIndexDuration = (int)Math.Round((1 - 2 * topFraction) * modifiedPitchMarkIndexSpacing);
int previousPitchMarkIndex = originalPitchMarkIndexList[iPitchMark - 1];
int previousTopEndIndex = previousPitchMarkIndex + (int)Math.Round(topFraction * modifiedPitchMarkIndexSpacing);
int startIndexPreviousPitchPeriod = previousTopEndIndex;
int currentPitchMarkIndex = originalPitchMarkIndexList[iPitchMark];
int currentTopStartIndex = currentPitchMarkIndex - (int)Math.Round(topFraction * modifiedPitchMarkIndexSpacing);
for (int ii = 0; ii < transitionIndexDuration; ii++)
{
double alpha = (double)ii / (double)(transitionIndexDuration - 1);
int previousPitchPeriodSampleIndex = previousTopEndIndex + ii;
int currentPitchPeriodSampleIndex = currentTopStartIndex - transitionIndexDuration + ii;
short newSample = (short)Math.Round(((1 - alpha) * sound.Samples[0][previousPitchPeriodSampleIndex] +
alpha * sound.Samples[0][currentPitchPeriodSampleIndex]));
newSamples.Add(newSample);
}
// Next, add samples around the top of the current pitch period:
if (iPitchMark < (originalPitchMarkIndexList.Count - 1))
{
int nextModifiedPitchMarkIndexSpacing = modifiedPitchMarkIndexSpacingList[iPitchMark];
int currentTopEndIndex = currentPitchMarkIndex + (int)Math.Round(topFraction * nextModifiedPitchMarkIndexSpacing);
for (int ii = currentTopStartIndex; ii < currentTopEndIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
}
else // Special treatment of the final pitch period:
{
int endIndex = sound.Samples[0].Count;
for (int ii = currentTopStartIndex; ii < endIndex; ii++)
{
newSamples.Add(sound.Samples[0][ii]);
}
}
}
// Finally, build the sound from the new samples:
WAVSound newSound = new WAVSound(sound.Name, sound.SampleRate, sound.NumberOfChannels, sound.BitsPerSample);
newSound.GenerateFromSamples(new List <List <short> >()
{
newSamples, newSamples
});
return(newSound);
}
public WAVSound GenerateSound(FormantSpecification formantSpecification)
{
double duration = formantSpecification.GetDuration();
int samplingFrequency = formantSpecification.SamplingFrequency;
double sampleTime = 1.0 / (double)samplingFrequency;
double nominalPulsePeriod = 1 / (double)formantSpecification.FundamentalFrequency;
int numberOfSamples = (int)Math.Round(duration / sampleTime);
// Generate voiced pulse train: This requires running through the sequence
// of formantSettings in the formantSpecification, in order to determine
// the pitch (and its inverse, the pulse period) at each time:
List <int> voicedPulseSpacingList = formantSpecification.GeneratePulseSpacingList();
List <double> voicedPulseTrain = new List <double>();
for (int ii = 0; ii < numberOfSamples; ii++)
{
voicedPulseTrain.Add(0);
} // To be adjusted below.
int sampleIndex = 0;
while (sampleIndex < voicedPulseTrain.Count)
{
voicedPulseTrain[sampleIndex] = 1.0;
sampleIndex += voicedPulseSpacingList[sampleIndex];
}
// Generate unvoiced pulse train:
if (randomNumberGenerator == null)
{
randomNumberGenerator = new Random();
}
// if (gaussian == null) { gaussian = new GaussianDistribution(0, 0.05, -1); }
List <double> unvoicedPulseTrain = new List <double>();
for (int ii = 0; ii < numberOfSamples; ii++)
{
unvoicedPulseTrain.Add(0);
if (randomNumberGenerator.NextDouble() < 0.5)
{
unvoicedPulseTrain[ii] = -WHITE_NOISE_LEVEL + 2 * WHITE_NOISE_LEVEL * randomNumberGenerator.NextDouble(); // gaussian.GetSample();
}
}
// Set up sinusoids:
int numberOfSinusoids = formantSpecification.GetNumberOfSinusoids();
List <DampedSinusoid> sinusoidList = new List <DampedSinusoid>();
for (int iSinusoid = 0; iSinusoid < numberOfSinusoids; iSinusoid++)
{
DampedSinusoid sinusoid = new DampedSinusoid(samplingFrequency);
sinusoidList.Add(sinusoid);
}
// Prepare for storing pitch:
if (storePitch)
{
// pitchList = new List<double>();
timePitchPeriodList = new List <List <double> >();
}
// Generate the relative amplitude list: Must be done separately, to handle
// transitions:
List <double> relativeAmplitudeList = formantSpecification.GenerateRelativeAmplitudeList();
// Generate the unscaled samples:
List <double> unscaledSampleList = new List <double>();
double time = 0;
sampleIndex = 0;
while (sampleIndex < numberOfSamples)
{
time = sampleIndex * sampleTime;
FormantSettings formantSettings = formantSpecification.GetInterpolatedSettings(sampleIndex);
for (int iSinusoid = 0; iSinusoid < sinusoidList.Count; iSinusoid++)
{
sinusoidList[iSinusoid].SetParameters(formantSettings.AmplitudeList[iSinusoid],
formantSettings.FrequencyList[iSinusoid],
formantSettings.BandwidthList[iSinusoid]);
}
double x = formantSettings.VoicedFraction * voicedPulseTrain[sampleIndex] +
(1 - formantSettings.VoicedFraction) * unvoicedPulseTrain[sampleIndex];
// 20170407
if (storePitch)
{
if (formantSettings.VoicedFraction > minimumVoicedFractionForPitch)
{
if (voicedPulseTrain[sampleIndex] != 0) // Define the pitch only at pulse spikes
{
double pitch = samplingFrequency / voicedPulseSpacingList[sampleIndex];
// pitchList.Add(pitch);
double pitchPeriod = voicedPulseSpacingList[sampleIndex] * sampleTime;
timePitchPeriodList.Add(new List <double>()
{
time, pitchPeriod
});
}
/* else { pitchList.Add(-1); }
* }
* else
* {
* pitchList.Add(-1); // < 0 => pitch not defined */
}
}
double deltaTime = formantSpecification.DeltaTimeList[sampleIndex];
double relativeAmplitude = relativeAmplitudeList[sampleIndex]; // formantSettings.GetRelativeAmplitude(deltaTime);
double sample = 0;
for (int iSinusoid = 0; iSinusoid < sinusoidList.Count; iSinusoid++)
{
sample += sinusoidList[iSinusoid].Next(x);
}
sample *= relativeAmplitude * volume;
unscaledSampleList.Add(sample);
sampleIndex++;
}
// Next generate the scaled samples
List <Int16> sampleList = new List <Int16>();
for (int ii = 0; ii < numberOfSamples; ii++)
{
if (unscaledSampleList[ii] > 1)
{
unscaledSampleList[ii] = 1;
}
else if (unscaledSampleList[ii] < -1)
{
unscaledSampleList[ii] = -1;
}
Int16 sample = (Int16)Math.Round(32767 * unscaledSampleList[ii]); // Some ugly hard-coding here...
sampleList.Add(sample);
}
List <List <Int16> > twoChannelSampleList = new List <List <Int16> >();
twoChannelSampleList.Add(sampleList);
WAVSound wavSound = new WAVSound("Test", samplingFrequency, 1, 16); // Some ugly hard-coding here...
wavSound.GenerateFromSamples(twoChannelSampleList);
return(wavSound);
}
