marytts.modules.synthesis.Voice Maven / Gradle / Ivy
The newest version!
/**
* Copyright 2000-2006 DFKI GmbH.
* All Rights Reserved. Use is subject to license terms.
*
* This file is part of MARY TTS.
*
* MARY TTS is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation, version 3 of the License.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with this program. If not, see allVoices = new TreeSet(new Comparator() {
public int compare(Voice v1, Voice v2) {
// Return negative number if v1 should be listed before v2
int desireDelta = v2.wantToBeDefault - v1.wantToBeDefault;
if (desireDelta != 0)
return desireDelta;
// same desire -- sort alphabetically
return v2.getName().compareTo(v1.getName());
}
});
private static Map defaultVoices = new HashMap();
protected static Logger logger = MaryUtils.getLogger("Voice");
private String voiceName;
private Locale locale;
private AudioFormat dbAudioFormat = null;
private WaveformSynthesizer synthesizer;
private Gender gender;
private int wantToBeDefault;
private AllophoneSet allophoneSet;
String preferredModulesClasses;
private Vector preferredModules;
private boolean vocalizationSupport;
private VocalizationSynthesizer vocalizationSynthesizer;
protected DirectedGraph durationGraph;
protected DirectedGraph f0Graph;
protected FeatureFileReader f0ContourFeatures;
protected Map acousticModels;
@Deprecated
public Voice(String name, Locale locale, AudioFormat dbAudioFormat, WaveformSynthesizer synthesizer, Gender gender)
throws MaryConfigurationException {
this.voiceName = name;
this.locale = locale;
this.dbAudioFormat = dbAudioFormat;
this.synthesizer = synthesizer;
this.gender = gender;
try {
init();
} catch (Exception n) {
throw new MaryConfigurationException("Cannot instantiate voice '" + voiceName + "'", n);
}
}
public Voice(String name, WaveformSynthesizer synthesizer) throws MaryConfigurationException {
this.voiceName = name;
this.synthesizer = synthesizer;
VoiceConfig config = MaryConfig.getVoiceConfig(voiceName);
if (config == null) {
throw new MaryConfigurationException("Trying to load config for voice '" + voiceName + "' but cannot find it.");
}
this.locale = config.getLocale();
int samplingRate = MaryProperties.getInteger("voice." + voiceName + ".samplingRate", 16000);
this.dbAudioFormat = new AudioFormat(AudioFormat.Encoding.PCM_SIGNED, samplingRate, // samples per second
16, // bits per sample
1, // mono
2, // nr. of bytes per frame
samplingRate, // nr. of frames per second
false);
this.gender = new Gender(MaryProperties.needProperty("voice." + voiceName + ".gender"));
try {
init();
} catch (Exception n) {
throw new MaryConfigurationException("Cannot instantiate voice '" + voiceName + "'", n);
}
}
/**
* @throws MaryConfigurationException
* MaryConfigurationException
* @throws NoSuchPropertyException
* NoSuchPropertyException
* @throws IOException
* IOException
*/
private void init() throws MaryConfigurationException, NoSuchPropertyException, IOException {
// Read settings from config file:
String header = "voice." + getName();
this.wantToBeDefault = MaryProperties.getInteger(header + ".wants.to.be.default", 0);
try {
allophoneSet = MaryRuntimeUtils.needAllophoneSet(header + ".allophoneset");
} catch (MaryConfigurationException e) {
// no allophone set for voice, try for locale
try {
allophoneSet = MaryRuntimeUtils.needAllophoneSet(MaryProperties.localePrefix(getLocale()) + ".allophoneset");
} catch (MaryConfigurationException e2) {
throw new MaryConfigurationException("No allophone set specified -- neither for voice '" + getName()
+ "' nor for locale '" + getLocale() + "'", e2);
}
}
preferredModulesClasses = MaryProperties.getProperty(header + ".preferredModules");
String lexiconClass = MaryProperties.getProperty(header + ".lexiconClass");
String lexiconName = MaryProperties.getProperty(header + ".lexicon");
vocalizationSupport = MaryProperties.getBoolean(header + ".vocalizationSupport", false);
if (vocalizationSupport) {
vocalizationSynthesizer = new VocalizationSynthesizer(this);
}
loadOldStyleProsodyModels(header);
loadAcousticModels(header);
// initialization of FeatureProcessorManager for this voice, if needed:
initFeatureProcessorManager();
}
@Deprecated
private void loadOldStyleProsodyModels(String header) throws MaryConfigurationException {
// see if there are any voice-specific duration and f0 models to load
durationGraph = null;
String durationGraphFile = MaryProperties.getFilename(header + ".duration.cart");
if (durationGraphFile != null) {
logger.debug("...loading duration graph...");
try {
durationGraph = (new DirectedGraphReader()).load(durationGraphFile);
} catch (IOException e) {
throw new MaryConfigurationException("Cannot load duration graph file '" + durationGraphFile + "'", e);
}
}
f0Graph = null;
String f0GraphFile = MaryProperties.getFilename(header + ".f0.graph");
if (f0GraphFile != null) {
logger.debug("...loading f0 contour graph...");
try {
f0Graph = (new DirectedGraphReader()).load(f0GraphFile);
// If we have the graph, we need the contour:
String f0ContourFile = MaryProperties.needFilename(header + ".f0.contours");
f0ContourFeatures = new FeatureFileReader(f0ContourFile);
} catch (IOException e) {
throw new MaryConfigurationException("Cannot load f0 contour graph file '" + f0GraphFile + "'", e);
}
}
}
/**
* Load a flexibly configurable list of acoustic models as specified in the config file.
*
* @param header
* header
* @throws MaryConfigurationException
* MaryConfigurationException
* @throws NoSuchPropertyException
* NoSuchPropertyException
* @throws IOException
* IOException
*/
private void loadAcousticModels(String header) throws MaryConfigurationException, NoSuchPropertyException, IOException {
// The feature processor manager that all acoustic models will use to predict their acoustics:
FeatureProcessorManager symbolicFPM = FeatureRegistry.determineBestFeatureProcessorManager(getLocale());
// Acoustic models:
String acousticModelsString = MaryProperties.getProperty(header + ".acousticModels");
if (acousticModelsString != null) {
acousticModels = new HashMap();
// add boundary "model" (which could of course be overwritten by appropriate properties in voice config):
acousticModels.put("boundary", new BoundaryModel(symbolicFPM, voiceName, null, "duration", null, null, null,
"boundaries"));
StringTokenizer acousticModelStrings = new StringTokenizer(acousticModelsString);
do {
String modelName = acousticModelStrings.nextToken();
// get more properties from voice config, depending on the model name:
String modelType = MaryProperties.needProperty(header + "." + modelName + ".model");
InputStream modelDataStream = MaryProperties.getStream(header + "." + modelName + ".data"); // not used for hmm
// models
String modelAttributeName = MaryProperties.needProperty(header + "." + modelName + ".attribute");
// the following are null if not defined; this is handled in the Model constructor:
String modelAttributeFormat = MaryProperties.getProperty(header + "." + modelName + ".attribute.format");
String modelFeatureName = MaryProperties.getProperty(header + "." + modelName + ".feature");
String modelPredictFrom = MaryProperties.getProperty(header + "." + modelName + ".predictFrom");
String modelApplyTo = MaryProperties.getProperty(header + "." + modelName + ".applyTo");
// consult the ModelType enum to find appropriate Model subclass...
ModelType possibleModelTypes = ModelType.fromString(modelType);
// if modelType is not in ModelType.values(), we don't know how to handle it:
if (possibleModelTypes == null) {
throw new MaryConfigurationException("Cannot handle unknown model type: " + modelType);
}
// ...and instantiate it in a switch statement:
Model model = null;
try {
switch (possibleModelTypes) {
case CART:
model = new CARTModel(symbolicFPM, voiceName, modelDataStream, modelAttributeName, modelAttributeFormat,
modelFeatureName, modelPredictFrom, modelApplyTo);
break;
case SOP:
model = new SoPModel(symbolicFPM, voiceName, modelDataStream, modelAttributeName, modelAttributeFormat,
modelFeatureName, modelPredictFrom, modelApplyTo);
break;
case HMM:
// if we already have a HMM duration or F0 model, and if this is the other of the two, and if so,
// and they use the same dataFile, then let them be the same instance:
// if this is the case set the boolean variable predictDurAndF0 to true in HMMModel
if (getDurationModel() != null && getDurationModel() instanceof HMMModel
&& modelName.equalsIgnoreCase("F0") && voiceName.equals(getDurationModel().getVoiceName())) {
model = getDurationModel();
((HMMModel) model).setPredictDurAndF0(true);
} else if (getF0Model() != null && getF0Model() instanceof HMMModel
&& modelName.equalsIgnoreCase("duration") && voiceName.equals(getF0Model().getVoiceName())) {
model = getF0Model();
((HMMModel) model).setPredictDurAndF0(true);
} else {
model = new HMMModel(symbolicFPM, voiceName, modelDataStream, modelAttributeName,
modelAttributeFormat, modelFeatureName, modelPredictFrom, modelApplyTo);
}
break;
}
} catch (Throwable t) {
throw new MaryConfigurationException("Cannot instantiate model '" + modelName + "' of type '" + modelType
+ "' from '" + MaryProperties.getProperty(header + "." + modelName + ".data") + "'", t);
}
// if we got this far, model should not be null:
assert model != null;
// put the model in the Model Map:
acousticModels.put(modelName, model);
} while (acousticModelStrings.hasMoreTokens());
}
}
/**
* Try to determine a feature processor manager. This will look for the voice-specific config setting
* voice.(voicename).featuremanager. If a feature processor manager is found, it is initialised and entered into
* the {@link marytts.features.FeatureRegistry}.
*
* @throws MaryConfigurationException
* if the feature processor manager cannot be initialised.
*/
private void initFeatureProcessorManager() throws MaryConfigurationException {
FeatureProcessorManager featMgr = null;
// Any feature processor manager settings in the config file?
String keyVoiceFeatMgr = "voice." + getName() + ".featuremanager";
String featMgrClass = MaryProperties.getProperty(keyVoiceFeatMgr);
if (featMgrClass != null) {
try {
featMgr = (FeatureProcessorManager) Class.forName(featMgrClass).newInstance();
} catch (Exception e) {
throw new MaryConfigurationException("Cannot initialise voice-specific FeatureProcessorManager " + featMgrClass
+ " from config file", e);
}
} else if (getOtherModels() != null) {
// Only if there is no feature manager setting in the config file,
// we consider creating one from the acoustic features;
// We need to do this only if we have any "other" acoustic models, beyond duration and F0:
FeatureProcessorManager genericFPM = FeatureRegistry.determineBestFeatureProcessorManager(locale);
// We attempt to create an FPM with same class as genericFPM via the Constructor FPM(Voice):
Class fpmClass = genericFPM.getClass();
try {
Constructor fpmVoiceConstructor = fpmClass.getConstructor(Voice.class);
featMgr = fpmVoiceConstructor.newInstance(this);
} catch (NoSuchMethodException nsme) {
throw new MaryConfigurationException("Cannot initialise voice-specific FeatureProcessorManager: Class "
+ fpmClass.getName() + " has no constructor " + fpmClass.getSimpleName() + "(Voice)");
} catch (Exception e) {
throw new MaryConfigurationException("Cannot initialise voice-specific FeatureProcessorManager", e);
}
}
// register the FeatureProcessorManager for this Voice:
if (featMgr != null) {
FeatureRegistry.setFeatureProcessorManager(this, featMgr);
}
}
/**
* Get the allophone set associated with this voice.
*
* @return allophoneSet
*/
public AllophoneSet getAllophoneSet() {
return allophoneSet;
}
/**
* Get the Allophone set for the given phone symbol.
*
* @param phoneSymbol
* phoneSymbol
* @return an Allophone object if phoneSymbol is a known phone symbol in the voice's AllophoneSet.
* @deprecated use {@link AllophoneSet#getAllophone(String)} directly instead
*/
@Deprecated
public Allophone getAllophone(String phoneSymbol) {
return allophoneSet.getAllophone(phoneSymbol);
}
public synchronized Vector getPreferredModulesAcceptingType(MaryDataType type) {
if (preferredModules == null && preferredModulesClasses != null) {
// need to initialise the list of modules
preferredModules = new Vector();
StringTokenizer st = new StringTokenizer(preferredModulesClasses);
while (st.hasMoreTokens()) {
String moduleInfo = st.nextToken();
try {
MaryModule mm = null;
if (!moduleInfo.contains("(")) { // no constructor info
mm = ModuleRegistry.getModule(Class.forName(moduleInfo));
}
if (mm == null) {
// need to create our own:
logger.warn("Module "
+ moduleInfo
+ " is not in the standard list of modules -- will start our own, but will not be able to shut it down at the end.");
mm = ModuleRegistry.instantiateModule(moduleInfo);
mm.startup();
}
preferredModules.add(mm);
} catch (Exception e) {
logger.warn("Cannot initialise preferred module " + moduleInfo + " for voice " + getName() + " -- skipping.",
e);
}
}
}
if (preferredModules != null) {
Vector v = new Vector();
for (Iterator it = preferredModules.iterator(); it.hasNext();) {
MaryModule m = (MaryModule) it.next();
if (m.inputType().equals(type)) {
v.add(m);
}
}
if (v.size() > 0)
return v;
else
return null;
}
return null;
}
public boolean hasName(String aName) {
return voiceName.equals(aName);
}
/**
* Return the name of this voice. If the voice has several possible names, the first one is returned.
*
* @return voiceName
*/
public String getName() {
return voiceName;
}
/** Returns the return value of getName(). */
public String toString() {
return getName();
}
public Locale getLocale() {
return locale;
}
public AudioFormat dbAudioFormat() {
return dbAudioFormat;
}
public WaveformSynthesizer synthesizer() {
return synthesizer;
}
public Gender gender() {
return gender;
}
public boolean hasVocalizationSupport() {
return vocalizationSupport;
}
public VocalizationSynthesizer getVocalizationSynthesizer() {
return vocalizationSynthesizer;
}
/**
* Get any styles supported by this voice.
*
* @return an array of style names supported by this voice, or null if styles are not supported.
*/
public String[] getStyles() {
// TODO: read from config file
if (voiceName.equals("dfki-pavoque-styles")) {
return new String[] { "neutral", "poker", "happy", "angry", "sad" };
}
return null;
}
/**
* Synthesize a list of tokens and boundaries with the waveform synthesizer providing this voice.
*
* @param tokensAndBoundaries
* tokensAndBoundaries
* @param outputParams
* outputParams
* @throws SynthesisException
* SynthesisException
* @return f0ContourFeatures
*/
public AudioInputStream synthesize(List tokensAndBoundaries, String outputParams) throws SynthesisException {
return synthesizer.synthesize(tokensAndBoundaries, this, outputParams);
}
public DirectedGraph getDurationGraph() {
return durationGraph;
}
public DirectedGraph getF0Graph() {
return f0Graph;
}
public FeatureFileReader getF0ContourFeatures() {
return f0ContourFeatures;
}
// Several getters for acoustic models, returning null if undefined:
/**
* Get the acoustic models defined for this voice.
*
* @return a Map mapping model names to models, or null if there are no such models.
*/
public Map getAcousticModels() {
return acousticModels;
}
/**
* Get the duration model for this voice.
*
* @return the model, or null if no such model is defined.
*/
public Model getDurationModel() {
if (acousticModels == null) {
return null;
}
return acousticModels.get("duration");
}
/**
* Get the F0 model for this voice.
*
* @return the model, or null if no such model is defined.
*/
public Model getF0Model() {
if (acousticModels == null) {
return null;
}
return acousticModels.get("F0");
}
/**
* Get the boundary duration model for this voice.
*
* @return the model, or null if no such model is defined.
*/
public Model getBoundaryModel() {
if (acousticModels == null) {
return null;
}
return acousticModels.get("boundary");
}
/**
* Return any "other" acoustic models that we have. Other models are acoustic models beyond duration, F0 and boundary.
*
* @return a Map mapping the model name to the model, or null if no other models exist.
*/
public Map getOtherModels() {
if (acousticModels == null) {
return null;
}
Map otherModels = new HashMap();
for (String modelName : acousticModels.keySet()) {
// ignore critical Models that have their own getters:
if (!modelName.equals("duration") && !modelName.equals("F0") && !modelName.equals("boundary")) {
otherModels.put(modelName, acousticModels.get(modelName));
}
}
if (otherModels.size() == 0) {
return null;
}
return otherModels;
}
// //////// static stuff //////////
/**
* Register the given voice. It will be contained in the list of available voices returned by any subsequent calls to
* getAvailableVoices(). If the voice has the highest value of wantToBeDefault for its locale it will be
* registered as the default voice for its locale. This value is set in the config file setting
* voice.(name).want.to.be.default.voice.
*
* @param voice
* voicwe
*/
public static void registerVoice(Voice voice) {
if (voice == null)
throw new NullPointerException("Cannot register null voice.");
if (!allVoices.contains(voice)) {
logger.info("Registering voice `" + voice.getName() + "': " + voice.gender() + ", locale " + voice.getLocale());
allVoices.add(voice);
}
checkIfDefaultVoice(voice);
}
/**
* Check if this voice should be registered as default.
*
* @param voice
* voice
*/
private static void checkIfDefaultVoice(Voice voice) {
Locale locale = voice.getLocale();
Voice currentDefault = defaultVoices.get(locale);
if (currentDefault == null || currentDefault.wantToBeDefault < voice.wantToBeDefault) {
logger.info("New default voice for locale " + locale + ": " + voice.getName() + " (desire " + voice.wantToBeDefault
+ ")");
defaultVoices.put(locale, voice);
}
}
/**
* Get the voice with the given name, or null if there is no voice with that name.
*
* @param name
* name
* @return v if it has name
*/
public static Voice getVoice(String name) {
for (Iterator it = allVoices.iterator(); it.hasNext();) {
Voice v = it.next();
if (v.hasName(name))
return v;
}
// Interpolating voices are created as needed:
if (InterpolatingVoice.isInterpolatingVoiceName(name)) {
InterpolatingSynthesizer interpolatingSynthesizer = null;
for (Iterator it = allVoices.iterator(); it.hasNext();) {
Voice v = it.next();
if (v instanceof InterpolatingVoice) {
interpolatingSynthesizer = (InterpolatingSynthesizer) v.synthesizer();
break;
}
}
if (interpolatingSynthesizer == null)
return null;
try {
Voice v = new InterpolatingVoice(interpolatingSynthesizer, name);
registerVoice(v);
return v;
} catch (Exception e) {
logger.warn("Could not create Interpolating voice:", e);
return null;
}
}
return null; // no such voice found
}
/**
* Get the list of all available voices. The iterator of the collection returned will return the voices in decreasing order of
* their "wantToBeDefault" value.
*
* @return Collections.unmodifiableSet(allVoices)
*/
public static Collection getAvailableVoices() {
return Collections.unmodifiableSet(allVoices);
}
/**
* Get the list of all available voices for a given locale. The iterator of the collection returned will return the voices in
* decreasing order of their "wantToBeDefault" value.
*
* @param locale
* locale
* @return a collection of Voice objects, or an empty collection if no voice is available for the given locale.
*/
public static Collection getAvailableVoices(Locale locale) {
ArrayList list = new ArrayList();
for (Voice v : allVoices) {
if (MaryUtils.subsumes(locale, v.getLocale())) {
list.add(v);
}
}
return list;
}
/**
* Get the list of all available voices for a given waveform synthesizer. The iterator of the collection returned will return
* the voices in decreasing order of their "wantToBeDefault" value.
*
* @param synth
* synth
* @return a collection of Voice objects, or an empty collection if no voice is available for the given waveform synthesizer.
*/
public static Collection getAvailableVoices(WaveformSynthesizer synth) {
if (synth == null) {
throw new NullPointerException("Got null WaveformSynthesizer");
}
ArrayList list = new ArrayList();
for (Voice v : allVoices) {
if (synth.equals(v.synthesizer())) {
list.add(v);
}
}
return list;
}
/**
* Get the list of all available voices for a given waveform synthesizer and locale. The iterator of the collection returned
* will return the voices in decreasing order of their "wantToBeDefault" value.
*
* @param synth
* synth
* @param locale
* locale
* @return a collection of Voice objects, or an empty collection if no voice is available for the given locale.
*/
public static Collection getAvailableVoices(WaveformSynthesizer synth, Locale locale) {
ArrayList list = new ArrayList();
for (Voice v : allVoices) {
if (v.synthesizer().equals(synth) && MaryUtils.subsumes(locale, v.getLocale())) {
list.add(v);
}
}
return list;
}
public static Voice getVoice(Locale locale, Gender gender) {
for (Voice v : allVoices) {
if (MaryUtils.subsumes(locale, v.getLocale()) && v.gender().equals(gender))
return v;
}
return null; // no such voice found
}
public static Voice getVoice(Element voiceElement) {
if (voiceElement == null || !voiceElement.getTagName().equals(MaryXML.VOICE)) {
return null;
}
Voice v = null;
// Try to get the voice by name:
String voiceName = voiceElement.getAttribute("name");
if (!voiceName.equals("")) {
v = Voice.getVoice(voiceName);
}
// Now if that didn't work, try getting a voice by gender:
if (v == null) {
String voiceGender = voiceElement.getAttribute("gender");
// Try to get the locale for the voice Element.
// Trust that the locale is encoded in the document root element.
Locale locale = MaryUtils
.string2locale(voiceElement.getOwnerDocument().getDocumentElement().getAttribute("xml:lang"));
if (locale == null) {
locale = Locale.GERMAN;
}
v = Voice.getVoice(locale, new Gender(voiceGender));
}
return v;
}
public static Voice getDefaultVoice(Locale locale) {
Voice v = defaultVoices.get(locale);
if (v == null)
v = getVoice(locale, FEMALE);
if (v == null)
v = getVoice(locale, MALE);
if (v == null)
logger.debug("Could not find default voice for locale " + locale);
return v;
}
public static Voice getSuitableVoice(MaryData d) {
Locale docLocale = d.getLocale();
if (docLocale == null && d.getType().isXMLType() && d.getDocument() != null
&& d.getDocument().getDocumentElement().hasAttribute("xml:lang")) {
docLocale = MaryUtils.string2locale(d.getDocument().getDocumentElement().getAttribute("xml:lang"));
}
Voice guessedVoice = null;
if (docLocale != null) {
guessedVoice = Voice.getDefaultVoice(docLocale);
} else {
// get any voice
if (allVoices.size() != 0)
guessedVoice = (Voice) allVoices.iterator().next();
}
if (guessedVoice != null)
logger.debug("Guessing default voice `" + guessedVoice.getName() + "'");
else
logger.debug("Couldn't find any voice at all");
return guessedVoice;
}
public static class Gender {
String name;
public Gender(String name) {
this.name = name;
}
public String toString() {
return name;
}
public boolean equals(Gender other) {
return other.toString().equals(name);
}
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy