public void append(MaryData md) {
if (md == null) throw new NullPointerException("Received null marydata");
if (!md.getType().equals(this.getType()))
throw new IllegalArgumentException(
"Cannot append mary data of type `"
+ md.getType().name()
+ "' to mary data of type `"
+ this.getType().name()
+ "'");
if (getType().isXMLType()) {
NodeList kids = md.getDocument().getDocumentElement().getChildNodes();
logger.debug("Appending " + kids.getLength() + " nodes to MaryXML structure");
Element docEl = this.getDocument().getDocumentElement();
for (int i = 0; i < kids.getLength(); i++) {
docEl.appendChild(this.getDocument().importNode(kids.item(i), true));
}
} else if (getType().isTextType()) {
// Attention: XML type is a text type!
if (this.plainText == null) {
this.plainText = md.getPlainText();
} else {
this.plainText = this.plainText + "\n\n" + md.getPlainText();
}
} else if (getType().equals(MaryDataType.get("AUDIO"))) {
appendAudio(md.getAudio());
} else {
throw new UnsupportedOperationException(
"Cannot append two mary data items of type `" + getType() + "'");
}
}
public MaryData process(MaryData d) throws Exception {
Document doc = d.getDocument();
NodeIterator it = MaryDomUtils.createNodeIterator(doc, doc, MaryXML.TOKEN);
Element t = null;
while ((t = (Element) it.nextNode()) != null) {
String text;
// Do not touch tokens for which a transcription is already
// given (exception: transcription contains a '*' character:
if (t.hasAttribute("ph") && !t.getAttribute("ph").contains("*")) {
continue;
}
if (t.hasAttribute("sounds_like")) text = t.getAttribute("sounds_like");
else text = MaryDomUtils.tokenText(t);
String pos = null;
// use part-of-speech if available
if (t.hasAttribute("pos")) {
pos = t.getAttribute("pos");
}
if (text != null
&& !text.equals("")
&& (pos == null || !pos.startsWith("$") /*punctuation*/)) {
// If text consists of several parts (e.g., because that was
// inserted into the sounds_like attribute), each part
// is transcribed separately.
StringBuilder ph = new StringBuilder();
String g2pMethod = null;
StringTokenizer st = new StringTokenizer(text, " -");
while (st.hasMoreTokens()) {
String graph = st.nextToken();
StringBuilder helper = new StringBuilder();
String phon = phonemise(graph, pos, helper);
if (ph.length() == 0) { // first part
// The g2pMethod of the combined beast is
// the g2pMethod of the first constituant.
g2pMethod = helper.toString();
ph.append(phon);
} else { // following parts
ph.append(" - ");
// Reduce primary to secondary stress:
ph.append(phon.replace('\'', ','));
}
}
if (ph != null && ph.length() > 0) {
setPh(t, ph.toString());
t.setAttribute("g2p_method", g2pMethod);
}
}
}
MaryData result = new MaryData(outputType(), d.getLocale());
result.setDocument(doc);
return result;
}
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;
}
public void computeFeaturesFor(String basename) throws IOException, Exception {
String text;
Locale localVoice;
localVoice = MaryUtils.string2locale(locale);
// First, test if there is a corresponding .rawmaryxml file in textdir:
File rawmaryxmlFile =
new File(db.getProp(db.MARYXMLDIR) + basename + db.getProp(db.MARYXMLEXT));
if (rawmaryxmlFile.exists()) {
text = FileUtils.getFileAsString(rawmaryxmlFile, "UTF-8");
} else {
text =
getMaryXMLHeaderWithInitialBoundary(locale)
+ FileUtils.getFileAsString(
new File(db.getProp(db.TEXTDIR) + basename + db.getProp(db.TEXTEXT)), "UTF-8")
+ "</maryxml>";
}
File pfeatFile = new File(unitfeatureDir, basename + featsExt);
OutputStream os = new BufferedOutputStream(new FileOutputStream(pfeatFile));
MaryClient maryClient = getMaryClient();
/*Vector voices = maryClient.getVoices(localVoice);
MaryClient.Voice defaultVoice = (MaryClient.Voice) voices.firstElement();
String voiceName = defaultVoice.name();*/
// maryClient.process(text, maryInputType, maryOutputType, null, null, os);
maryClient.process(text, maryInputType, maryOutputType, locale, null, "slt-arctic", os);
// maryClient.process(text, maryInputType, maryOutputType, null, "slt-arctic", os, timeout);
// maryClient.getOutputDataTypes().size()
// MaryData result = new MaryData(os);
os.flush();
os.close();
// System.out.println(" TO STRING: "+new FileReader(pfeatFile).toString());
// BufferedReader bfr = new BufferedReader(new FileReader(pfeatFile));
String line;
MaryData d = new MaryData(MaryDataType.get("PHONEMISED_EN"), Locale.US);
// d.readFrom(new ByteArrayInputStream(os.toByteArray()));
d.readFrom(new FileReader(pfeatFile));
// MaryData d = new MaryData(pfeatFile);
Document doc = d.getDocument();
// Document acoustparams = d.getDocument();
// NodeIterator it = ((DocumentTraversal)acoustparams).createNodeIterator(acoustparams,
// NodeFilter.SHOW_ELEMENT,new NameNodeFilter(new String[]{MaryXML.TOKEN,
// MaryXML.BOUNDARY}),false);
NodeIterator it =
((DocumentTraversal) doc)
.createNodeIterator(
doc, NodeFilter.SHOW_ELEMENT, new NameNodeFilter(MaryXML.TOKEN), false);
Element t = null;
while ((t = (Element) it.nextNode()) != null) {
if (t.hasAttribute("g2p_method")) {
String g2p = t.getAttribute("g2p_method");
String nodeText = t.getTextContent().trim();
if (g2p.equals("rules")) { // && nodeText.equals("!")){
System.out.print(basename + " ----> " + nodeText);
if (bnl.contains(basename)) bnl.remove(basename);
System.out.println(" SO removing basename: " + basename);
}
// System.out.println("G2P:"+t.getAttribute("g2p_method"));
// System.out.println("Text:"+t.getTextContent());
}
}
/*while((line =bfr.readLine()) != null){
//boolean b = m.matches();
if(Pattern.matches("rules", line))
System.out.println(basename + " LINE ---> " + line);
}*/
// System.out.println(" TO STRING: "+line);
}