public String toString() {
return Objects.toStringHelper(this)
.add("type", getType())
.add("locale", getLocale())
.add("output parameters", getOutputParams())
.add("data", getData())
.add("document", DomUtils.serializeToString(getDocument()))
.add("validating", getValidating())
.add("plain text", getPlainText())
.add("audio", getAudio())
.add("audio file format", getAudioFileFormat())
.toString();
}
/**
* 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");
}
}
private void updateVoiceDescription(File rootDir, File packageFile)
throws MalformedURLException, ParserConfigurationException, MaryConfigurationException,
IOException {
logger.debug("writing new voice description...");
voiceDescription.setVersion(Version.specificationVersion());
voiceDescription.setDependsVersion(Version.specificationVersion());
voiceDescription.setPackageFilename(packageFile.getName());
voiceDescription.setPackageMD5Sum(computeMD5(packageFile));
voiceDescription.setPackageSize((int) packageFile.length());
voiceDescription.removeAllLocations();
voiceDescription.addLocation(
URI.create("http://mary.dfki.de/download/" + Version.specificationVersion() + "/").toURL());
Document doc = voiceDescription.createComponentXML();
File newVoiceDescriptionFile = new File(rootDir, getFilenamePrefix() + "-component.xml");
DomUtils.document2File(doc, newVoiceDescriptionFile);
logger.debug("... created " + newVoiceDescriptionFile.getPath());
}
/**
* Preselect a set of candidates that could be used to realise the given target.
*
* @param target a Target object representing an optimal unit
* @return an <span style="color:red;">unsorted</span> ArrayList of ViterbiCandidates, each
* containing the (same) target and a (different) Unit object
*/
public List<ViterbiCandidate> getCandidates(Target target) {
// BEGIN blacklisting
// The point of this is to get the value of the "blacklist" attribute in the first child element
// of the MaryXML
// and store it in the blacklist String variable.
// This code seems rather inelegant; perhaps there is a better way to access the MaryXML from
// this method?
String blacklist = "";
String unitBasename =
"This must never be null or the empty string!"; // otherwise candidate selection fails!
Element targetElement = target.getMaryxmlElement();
blacklist = DomUtils.getAttributeFromClosestAncestorOfAnyKind(targetElement, "blacklist");
// END blacklisting
// logger.debug("Looking for candidates in cart "+target.getName());
// get the cart tree and extract the candidates
int[] clist = (int[]) preselectionCART.interpret(target, backtrace);
logger.debug("For target " + target + ", selected " + clist.length + " units");
// Now, clist is an array of unit indexes.
List<ViterbiCandidate> candidates = new ArrayList<ViterbiCandidate>();
for (int i = 0; i < clist.length; i++) {
// The target is the same for all these candidates in the queue
// remember the actual unit:
Unit unit = unitReader.getUnit(clist[i]);
candidates.add(new ViterbiCandidate(target, unit, targetCostFunction));
}
// Blacklisting without crazy performance drop:
// just remove candidates again if their basenames are blacklisted
java.util.Iterator<ViterbiCandidate> candIt = candidates.iterator();
while (candIt.hasNext()) {
ViterbiCandidate candidate = candIt.next();
unitBasename = getFilename(candidate.getUnit());
if (blacklist.contains(unitBasename)) {
candIt.remove();
}
}
return candidates;
}
public MaryData process(MaryData d) throws Exception {
// prevUnitIndex;
// numberOfConsecutiveUnits;
// basenameDuration;
// phoneTier;
// PraatIntervalTier unitTier;
// PraatIntervalTier sourceTier;
// sourceInterval;
Document doc = d.getDocument();
// initialize various variables:
Double duration = 0.0;
String phone = null;
// initialize some class variables:
PraatIntervalTier phoneTier = new PraatIntervalTier("phones");
Double basenameDuration = 0.0;
int prevUnitIndex = Integer.MIN_VALUE;
int numberOfConsecutiveUnits = 0; // counter to track consecutive units
PraatInterval sourceInterval = new PraatInterval(basenameDuration);
// until we have a robust way of checking the voice type, just initialize unit and source tiers
// anyway:
PraatIntervalTier unitTier = new PraatIntervalTier("units");
PraatIntervalTier sourceTier = new PraatIntervalTier("sources");
// prepare to iterate only over the PHONE, SENTENCE, and BOUNDARY nodes in the MaryXML:
NodeIterator ni = DomUtils.createNodeIterator(doc, PHONE, BOUNDARY);
Element element;
// now iterate over these nodes:
while ((element = (Element) ni.nextNode()) != null) {
switch (element.getTagName()) { // <s>, <ph>, or <boundary> as specified above
case PHONE:
phone = element.getAttribute("p");
duration =
Integer.parseInt(element.getAttribute("d")) / 1000.0; // duration is always in ms
break;
case BOUNDARY:
phone =
"_"; // TODO: perhaps we should access TargetFeatureComputer.getPauseSymbol() instead
if (element.hasAttribute("duration")) {
duration =
Double.parseDouble(element.getAttribute("duration"))
/ 1000.0; // duration is always in ms
} else {
duration = 0.0; // HMM voices can have duration-less <boundary/> tags
}
break;
default:
logger.error(
"NodeIterator should not find an element of type " + element.getTagName() + " here!");
break;
}
PraatInterval phoneInterval = new PraatInterval(duration, phone);
// TODO: crude way of checking for unit selection voice; also, name of attribute could change!
if (element.hasAttribute("units")) {
// unitselectionProcessing(element, unitTier, prevUnitIndex, numberOfConsecutiveUnits,
// basenameDuration,
// sourceInterval, sourceTier);
String units = element.getAttribute("units");
String[] unitStrings = units.split("; "); // boundaries have only one unit string
boolean differentSource = false;
String basename = null;
String unitRange = null;
for (String unitString : unitStrings) {
// TODO verify that unit string matches "UNITNAME BASENAME UNITINDEX UNITDURATION"
String[] unitFields = unitString.split(" ");
String unitName = unitFields[0];
basename = unitFields[1];
int unitIndex = Integer.parseInt(unitFields[2]);
Double unitDuration = Double.parseDouble(unitFields[3]);
// units are straightforward, just like phones:
unitTier.appendInterval(new PraatInterval(unitDuration, unitString));
// unit source processing is a little more elaborate:
/*
* Note: the following assumes that consecutive selected units are ALWAYS from the same basename! That could
* change if basename boundaries are no longer marked by null units in the timeline.
*/
differentSource =
unitIndex
!= prevUnitIndex + 1; // is source unit from a different part of the timeline?;
if (differentSource) {
// reset primary variables:
numberOfConsecutiveUnits = 0;
basenameDuration = 0.0;
}
// increment/increase primary variables:
numberOfConsecutiveUnits++;
basenameDuration += unitDuration;
// construct unit index range string:
unitRange = Integer.toString(unitIndex - numberOfConsecutiveUnits + 1);
if (numberOfConsecutiveUnits > 1) {
unitRange = unitRange + "-" + unitIndex;
}
// append source intervals to source tier:
if (differentSource) {
sourceInterval = new PraatInterval(basenameDuration, basename + ": " + unitRange);
sourceTier.appendInterval(sourceInterval);
} else {
sourceInterval.setDuration(basenameDuration);
sourceInterval.setText(basename + ": " + unitRange);
}
prevUnitIndex = unitIndex;
}
// HACK: arbitrary threshold to detect end points in ms (in the case of diphone voice or
// boundary segment)
} else if (duration > 10) {
// TODO: there is still a bug somewhere regarding boundary durations with mbrola...
phoneInterval.setDuration(duration / 1000.0);
}
phoneTier.appendInterval(phoneInterval);
}
PraatTextGrid textGrid = new PraatTextGrid();
phoneTier.updateBoundaries(); // force full specification of timings
textGrid.appendTier(phoneTier);
// fragile way of checking whether this is a unit selection voice:
if (unitTier.getNumberOfIntervals() > 0) {
// complete and append unit and source tiers:
unitTier.updateBoundaries();
textGrid.appendTier(unitTier);
sourceTier.updateBoundaries();
textGrid.appendTier(sourceTier);
}
// return raw TextGrid as result:
MaryData result = new MaryData(getOutputType(), d.getLocale());
result.setPlainText(textGrid.toString());
return result;
}
