/**
* For the voice with the given name, return the list of vocalizations supported by this voice,
* one vocalization per line. These values can be used in the "name" attribute of the vocalization
* tag.
*
* @param voiceName
* @return the list of vocalizations, or the empty string if the voice does not support
* vocalizations.
*/
public static String getVocalizations(String voiceName) {
Voice v = Voice.getVoice(voiceName);
if (v == null || !v.hasVocalizationSupport()) {
return "";
}
VocalizationSynthesizer vs = v.getVocalizationSynthesizer();
assert vs != null;
String[] vocalizations = vs.listAvailableVocalizations();
assert vocalizations != null;
return StringUtils.toString(vocalizations);
}
/**
* Set the content data of this MaryData object from the given String. For XML data ({@link
* MaryDataType#isXMLType()}), parse the String representation of the data into a DOM tree.
*
* @param dataString string representation of the input data.
* @throws ParserConfigurationException ParserConfigurationException
* @throws IOException IOException
* @throws SAXException SAXException
* @throws IllegalArgumentException if this method is called for MaryDataTypes that are neither
* text nor XML.
*/
public void setData(String dataString)
throws ParserConfigurationException, SAXException, IOException {
// First, some data cleanup:
dataString = StringUtils.purgeNonBreakingSpaces(dataString);
// Now, deal with it.
if (type.isXMLType()) {
logger.debug(
"Parsing XML input (" + (doValidate ? "" : "non-") + "validating): " + dataString);
xmlDocument = DomUtils.parseDocument(dataString, doValidate);
} else if (type.isTextType()) {
logger.debug("Setting text input: " + dataString);
plainText = dataString;
} else {
throw new IllegalArgumentException("Cannot set data of type " + type + " from a string");
}
}
public static Vector<String> getDefaultVoiceExampleTexts() {
String defaultVoiceName = getDefaultVoiceName();
Vector<String> defaultVoiceExampleTexts = null;
defaultVoiceExampleTexts =
StringUtils.processVoiceExampleText(getVoiceExampleText(defaultVoiceName));
if (defaultVoiceExampleTexts == null) // Try for general domain
{
String str = getExampleText("TEXT", Voice.getVoice(defaultVoiceName).getLocale());
if (str != null && str.length() > 0) {
defaultVoiceExampleTexts = new Vector<String>();
defaultVoiceExampleTexts.add(str);
}
}
return defaultVoiceExampleTexts;
}
// Pseudo harmonics based noise generation for pseudo periods
public static double[] synthesize(
HntmSpeechSignal hnmSignal,
HntmAnalyzerParams analysisParams,
HntmSynthesizerParams synthesisParams,
String referenceFile) {
double[] noisePart = null;
int trackNoToExamine = 1;
int i, k, n;
double t; // Time in seconds
double tsik = 0.0; // Synthesis time in seconds
double tsikPlusOne = 0.0; // Synthesis time in seconds
double trackStartInSeconds, trackEndInSeconds;
// double lastPeriodInSeconds = 0.0;
int trackStartIndex, trackEndIndex;
double akt;
int numHarmonicsCurrentFrame, numHarmonicsPrevFrame, numHarmonicsNextFrame;
int harmonicIndexShiftPrev, harmonicIndexShiftCurrent, harmonicIndexShiftNext;
int maxNumHarmonics = 0;
for (i = 0; i < hnmSignal.frames.length; i++) {
if (hnmSignal.frames[i].maximumFrequencyOfVoicingInHz > 0.0f
&& hnmSignal.frames[i].n != null) {
numHarmonicsCurrentFrame =
(int) Math.floor(hnmSignal.samplingRateInHz / analysisParams.noiseF0InHz + 0.5);
numHarmonicsCurrentFrame = Math.max(0, numHarmonicsCurrentFrame);
if (numHarmonicsCurrentFrame > maxNumHarmonics) maxNumHarmonics = numHarmonicsCurrentFrame;
}
}
double aksi;
double aksiPlusOne;
float[] phasekis = null;
float phasekiPlusOne;
double ht;
float phasekt = 0.0f;
float phasekiEstimate = 0.0f;
float phasekiPlusOneEstimate = 0.0f;
int Mk;
boolean isPrevNoised, isNoised, isNextNoised;
boolean isTrackNoised, isNextTrackNoised, isPrevTrackNoised;
int outputLen =
SignalProcUtils.time2sample(
hnmSignal.originalDurationInSeconds, hnmSignal.samplingRateInHz);
noisePart =
new double
[outputLen]; // In fact, this should be prosody scaled length when you implement prosody
// modifications
Arrays.fill(noisePart, 0.0);
// Write separate tracks to output
double[][] noiseTracks = null;
if (maxNumHarmonics > 0) {
noiseTracks = new double[maxNumHarmonics][];
for (k = 0; k < maxNumHarmonics; k++) {
noiseTracks[k] = new double[outputLen];
Arrays.fill(noiseTracks[k], 0.0);
}
phasekis = new float[maxNumHarmonics];
for (k = 0; k < maxNumHarmonics; k++)
phasekis[k] = (float) (MathUtils.TWOPI * (Math.random() - 0.5));
}
//
int transitionLen =
SignalProcUtils.time2sample(
synthesisParams.unvoicedVoicedTrackTransitionInSeconds, hnmSignal.samplingRateInHz);
Window transitionWin = Window.get(Window.HAMMING, transitionLen * 2);
transitionWin.normalizePeakValue(1.0f);
double[] halfTransitionWinLeft = transitionWin.getCoeffsLeftHalf();
float halfFs = hnmSignal.samplingRateInHz;
for (i = 0; i < hnmSignal.frames.length; i++) {
isPrevNoised = false;
isNoised = false;
isNextNoised = false;
if (i > 0
&& hnmSignal.frames[i - 1].n != null
&& hnmSignal.frames[i - 1].maximumFrequencyOfVoicingInHz < halfFs
&& ((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i - 1].n).ceps != null)
isPrevNoised = true;
if (i > 0
&& hnmSignal.frames[i].n != null
&& hnmSignal.frames[i].maximumFrequencyOfVoicingInHz < halfFs
&& ((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i].n).ceps != null) isNoised = true;
if (i < hnmSignal.frames.length - 1
&& hnmSignal.frames[i + 1].maximumFrequencyOfVoicingInHz < halfFs
&& hnmSignal.frames[i + 1].n != null
&& ((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i + 1].n).ceps != null)
isNextNoised = true;
numHarmonicsPrevFrame = 0;
numHarmonicsCurrentFrame = 0;
numHarmonicsNextFrame = 0;
harmonicIndexShiftPrev = 0;
harmonicIndexShiftCurrent = 0;
harmonicIndexShiftNext = 0;
if (isPrevNoised) {
numHarmonicsPrevFrame =
(int)
Math.floor(
(hnmSignal.samplingRateInHz
- hnmSignal.frames[i - 1].maximumFrequencyOfVoicingInHz)
/ analysisParams.noiseF0InHz
+ 0.5);
numHarmonicsPrevFrame = Math.max(0, numHarmonicsPrevFrame);
harmonicIndexShiftPrev =
(int)
Math.floor(
hnmSignal.frames[i - 1].maximumFrequencyOfVoicingInHz
/ analysisParams.noiseF0InHz
+ 0.5);
harmonicIndexShiftPrev = Math.max(1, harmonicIndexShiftPrev);
}
if (isNoised) {
numHarmonicsCurrentFrame =
(int)
Math.floor(
(hnmSignal.samplingRateInHz - hnmSignal.frames[i].maximumFrequencyOfVoicingInHz)
/ analysisParams.noiseF0InHz
+ 0.5);
numHarmonicsCurrentFrame = Math.max(0, numHarmonicsCurrentFrame);
harmonicIndexShiftCurrent =
(int)
Math.floor(
hnmSignal.frames[i].maximumFrequencyOfVoicingInHz / analysisParams.noiseF0InHz
+ 0.5);
harmonicIndexShiftCurrent = Math.max(1, harmonicIndexShiftCurrent);
} else if (!isNoised && isNextNoised) {
numHarmonicsCurrentFrame =
(int)
Math.floor(
(hnmSignal.samplingRateInHz
- hnmSignal.frames[i + 1].maximumFrequencyOfVoicingInHz)
/ analysisParams.noiseF0InHz
+ 0.5);
numHarmonicsCurrentFrame = Math.max(0, numHarmonicsCurrentFrame);
harmonicIndexShiftCurrent =
(int)
Math.floor(
hnmSignal.frames[i + 1].maximumFrequencyOfVoicingInHz
/ analysisParams.noiseF0InHz
+ 0.5);
harmonicIndexShiftCurrent = Math.max(1, harmonicIndexShiftCurrent);
}
if (isNextNoised) {
numHarmonicsNextFrame =
(int)
Math.floor(
(hnmSignal.samplingRateInHz
- hnmSignal.frames[i + 1].maximumFrequencyOfVoicingInHz)
/ analysisParams.noiseF0InHz
+ 0.5);
numHarmonicsNextFrame = Math.max(0, numHarmonicsNextFrame);
harmonicIndexShiftNext =
(int)
Math.floor(
hnmSignal.frames[i + 1].maximumFrequencyOfVoicingInHz
/ analysisParams.noiseF0InHz
+ 0.5);
harmonicIndexShiftNext = Math.max(1, harmonicIndexShiftNext);
}
for (k = 0; k < numHarmonicsCurrentFrame; k++) {
aksi = 0.0;
aksiPlusOne = 0.0;
phasekiPlusOne = 0.0f;
isPrevTrackNoised = false;
isTrackNoised = false;
isNextTrackNoised = false;
if (i > 0 && hnmSignal.frames[i - 1].n != null && numHarmonicsPrevFrame > k)
isPrevTrackNoised = true;
if (hnmSignal.frames[i].n != null && numHarmonicsCurrentFrame > k) isTrackNoised = true;
if (i < hnmSignal.frames.length - 1
&& hnmSignal.frames[i + 1].n != null
&& numHarmonicsNextFrame > k) isNextTrackNoised = true;
tsik = hnmSignal.frames[i].tAnalysisInSeconds;
if (i == 0) trackStartInSeconds = 0.0;
else trackStartInSeconds = tsik;
if (i == hnmSignal.frames.length - 1) tsikPlusOne = hnmSignal.originalDurationInSeconds;
else tsikPlusOne = hnmSignal.frames[i + 1].tAnalysisInSeconds;
trackEndInSeconds = tsikPlusOne;
trackStartIndex =
SignalProcUtils.time2sample(trackStartInSeconds, hnmSignal.samplingRateInHz);
trackEndIndex = SignalProcUtils.time2sample(trackEndInSeconds, hnmSignal.samplingRateInHz);
if (isTrackNoised && trackEndIndex - trackStartIndex + 1 > 0) {
// Amplitudes
if (isTrackNoised) {
if (!analysisParams.useNoiseAmplitudesDirectly) {
if (analysisParams.regularizedCepstrumWarpingMethod
== RegularizedCepstrumEstimator.REGULARIZED_CEPSTRUM_WITH_PRE_BARK_WARPING)
aksi =
RegularizedPreWarpedCepstrumEstimator.cepstrum2linearSpectrumValue(
((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i].n).ceps,
(k + harmonicIndexShiftCurrent) * analysisParams.noiseF0InHz,
hnmSignal.samplingRateInHz);
else if (analysisParams.regularizedCepstrumWarpingMethod
== RegularizedCepstrumEstimator.REGULARIZED_CEPSTRUM_WITH_POST_MEL_WARPING)
aksi =
RegularizedPostWarpedCepstrumEstimator.cepstrum2linearSpectrumValue(
((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i].n).ceps,
(k + harmonicIndexShiftCurrent) * analysisParams.noiseF0InHz,
hnmSignal.samplingRateInHz);
} else {
if (k < ((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i].n).ceps.length)
aksi =
((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i].n)
.ceps[k]; // Use amplitudes directly without cepstrum method
else aksi = 0.0;
}
} else aksi = 0.0;
if (isNextTrackNoised) {
if (!analysisParams.useNoiseAmplitudesDirectly) {
if (analysisParams.regularizedCepstrumWarpingMethod
== RegularizedCepstrumEstimator.REGULARIZED_CEPSTRUM_WITH_PRE_BARK_WARPING)
aksiPlusOne =
RegularizedPreWarpedCepstrumEstimator.cepstrum2linearSpectrumValue(
((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i + 1].n).ceps,
(k + harmonicIndexShiftNext) * analysisParams.noiseF0InHz,
hnmSignal.samplingRateInHz);
else if (analysisParams.regularizedCepstrumWarpingMethod
== RegularizedCepstrumEstimator.REGULARIZED_CEPSTRUM_WITH_POST_MEL_WARPING)
aksiPlusOne =
RegularizedPostWarpedCepstrumEstimator.cepstrum2linearSpectrumValue(
((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i + 1].n).ceps,
(k + harmonicIndexShiftNext) * analysisParams.noiseF0InHz,
hnmSignal.samplingRateInHz);
} else {
if (k < ((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i + 1].n).ceps.length)
aksiPlusOne =
((FrameNoisePartPseudoHarmonic) hnmSignal.frames[i + 1].n)
.ceps[k]; // Use amplitudes directly without cepstrum method
else aksiPlusOne = 0.0;
}
} else aksiPlusOne = 0.0;
//
// Phases
phasekis[k] = (float) (MathUtils.TWOPI * (Math.random() - 0.5));
phasekiPlusOne =
(float)
(phasekis[k]
+ (k + harmonicIndexShiftCurrent)
* MathUtils.TWOPI
* analysisParams.noiseF0InHz
* (tsikPlusOne - tsik)); // Equation (3.55)
//
if (!isPrevTrackNoised) trackStartIndex = Math.max(0, trackStartIndex - transitionLen);
for (n = trackStartIndex; n <= Math.min(trackEndIndex, outputLen - 1); n++) {
t = SignalProcUtils.sample2time(n, hnmSignal.samplingRateInHz);
// if (t>=tsik && t<tsikPlusOne)
{
// Amplitude estimate
akt = MathUtils.interpolatedSample(tsik, t, tsikPlusOne, aksi, aksiPlusOne);
//
// Phase estimate
phasekt = (float) (phasekiPlusOne * (t - tsik) / (tsikPlusOne - tsik));
//
if (!isPrevTrackNoised && n - trackStartIndex < transitionLen)
noiseTracks[k][n] =
halfTransitionWinLeft[n - trackStartIndex] * akt * Math.cos(phasekt);
else noiseTracks[k][n] = akt * Math.cos(phasekt);
}
}
phasekis[k] = phasekiPlusOne;
}
}
}
for (k = 0; k < noiseTracks.length; k++) {
for (n = 0; n < noisePart.length; n++) noisePart[n] += noiseTracks[k][n];
}
// Write separate tracks to output
if (noiseTracks != null) {
for (k = 0; k < noiseTracks.length; k++) {
for (n = 0; n < noisePart.length; n++) noisePart[n] += noiseTracks[k][n];
}
if (referenceFile != null
&& FileUtils.exists(referenceFile)
&& synthesisParams.writeSeparateHarmonicTracksToOutputs) {
// Write separate tracks to output
AudioInputStream inputAudio = null;
try {
inputAudio = AudioSystem.getAudioInputStream(new File(referenceFile));
} catch (UnsupportedAudioFileException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (IOException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
if (inputAudio != null) {
// k=1;
for (k = 0; k < noiseTracks.length; k++) {
noiseTracks[k] = MathUtils.divide(noiseTracks[k], 32767.0);
DDSAudioInputStream outputAudio =
new DDSAudioInputStream(
new BufferedDoubleDataSource(noiseTracks[k]), inputAudio.getFormat());
String outFileName =
StringUtils.getFolderName(referenceFile)
+ "noiseTrack"
+ String.valueOf(k + 1)
+ ".wav";
try {
AudioSystem.write(outputAudio, AudioFileFormat.Type.WAVE, new File(outFileName));
} catch (IOException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
}
}
//
}
return noisePart;
}
