/**
* Portions Copyright 2006-2007 DFKI GmbH.
* Portions Copyright 2001 Sun Microsystems, Inc.
* Portions Copyright 1999-2001 Language Technologies Institute,
* Carnegie Mellon University.
* All Rights Reserved. Use is subject to license terms.
*
* Permission is hereby granted, free of charge, to use and distribute
* this software and its documentation without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of this work, and to
* permit persons to whom this work is furnished to do so, subject to
* the following conditions:
*
* 1. The code must retain the above copyright notice, this list of
* conditions and the following disclaimer.
* 2. Any modifications must be clearly marked as such.
* 3. Original authors' names are not deleted.
* 4. The authors' names are not used to endorse or promote products
* derived from this software without specific prior written
* permission.
*
* DFKI GMBH AND THE CONTRIBUTORS TO THIS WORK DISCLAIM ALL WARRANTIES WITH
* REGARD TO THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL DFKI GMBH NOR THE
* CONTRIBUTORS BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL
* DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
* PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
* ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF
* THIS SOFTWARE.
*/
package marytts.features;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.HashSet;
import java.util.Set;
import java.util.StringTokenizer;
import marytts.datatypes.MaryXML;
import marytts.unitselection.select.DiphoneTarget;
import marytts.unitselection.select.HalfPhoneTarget;
import marytts.unitselection.select.Target;
import marytts.util.dom.MaryDomUtils;
import marytts.util.string.ByteStringTranslator;
import org.w3c.dom.Document;
import org.w3c.dom.Element;
import org.w3c.dom.traversal.TreeWalker;
/**
* A collection of feature processors that operate on Target objects.
*
* @author schroed
*
*/
public class MaryGenericFeatureProcessors {
/**
* Navigate from a target to an item. Classes implementing this interface will retrieve meaningful items given the target.
*
* @author Marc Schröder
*/
public static interface TargetElementNavigator {
/**
* Given the target, retrieve an XML Element.
*
* @param target
* target
* @return an item selected according to this navigator, or null if there is no such item.
*/
public Element getElement(Target target);
}
/**
* Retrieve the segment belonging to this target.
*
* @author Marc Schröder
*
*/
public static class SegmentNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
return target.getMaryxmlElement();
}
}
/**
* Retrieve the segment preceding the segment which belongs to this target.
*
* @author Marc Schröder
*
*/
public static class PrevSegmentNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.PHONE, MaryXML.BOUNDARY);
tw.setCurrentNode(segment);
Element previous = (Element) tw.previousNode();
return previous;
}
}
/**
* Retrieve the segment two before the segment which belongs to this target.
*
* @author Marc Schröder
*
*/
public static class PrevPrevSegmentNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.PHONE, MaryXML.BOUNDARY);
tw.setCurrentNode(segment);
Element previous = (Element) tw.previousNode();
Element pp = (Element) tw.previousNode();
return pp;
}
}
/**
* Retrieve the segment following the segment which belongs to this target.
*
* @author Marc Schröder
*
*/
public static class NextSegmentNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.PHONE, MaryXML.BOUNDARY);
tw.setCurrentNode(segment);
Element next = (Element) tw.nextNode();
return next;
}
}
/**
* Retrieve the segment two after the segment which belongs to this target.
*
* @author Marc Schröder
*
*/
public static class NextNextSegmentNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.PHONE, MaryXML.BOUNDARY);
tw.setCurrentNode(segment);
Element next = (Element) tw.nextNode();
Element nn = (Element) tw.nextNode();
return nn;
}
}
/**
* Retrieve the first segment in the word to which this target belongs.
*
*/
public static class FirstSegmentInWordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.PHONE);
Element first = (Element) tw.firstChild();
if (first != null) {
assert first.getTagName().equals(MaryXML.PHONE) : "Unexpected tag name: expected " + MaryXML.PHONE + ", got "
+ first.getTagName();
}
return first;
}
}
/**
* Retrieve the last segment in the word to which this target belongs.
*
*/
public static class LastSegmentInWordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.PHONE);
Element last = (Element) tw.lastChild();
if (last != null) {
assert last.getTagName().equals(MaryXML.PHONE) : "Unexpected tag name: expected " + MaryXML.PHONE + ", got "
+ last.getTagName();
}
return last;
}
}
/**
* Retrieve the first syllable in the word to which this target belongs.
*
*/
public static class FirstSyllableInWordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.SYLLABLE);
Element first = (Element) tw.firstChild();
if (first != null) {
assert first.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE
+ ", got " + first.getTagName();
}
return first;
}
}
/**
* Retrieve the last syllable in the word to which this target belongs.
*
*/
public static class LastSyllableInWordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.SYLLABLE);
Element last = (Element) tw.lastChild();
if (last != null) {
assert last.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE
+ ", got " + last.getTagName();
}
return last;
}
}
/**
* Retrieve the syllable belonging to this target.
*
* @author Marc Schröder
*
*/
public static class SyllableNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
if (!segment.getTagName().equals(MaryXML.PHONE))
return null;
Element syllable = (Element) segment.getParentNode();
if (syllable != null) {
assert syllable.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE
+ ", got " + syllable.getTagName();
}
return syllable;
}
}
/**
* Retrieve the syllable before the syllable belonging to this target.
*
* @author Marc Schröder
*
*/
public static class PrevSyllableNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return null;
current = syllable;
} else { // boundary
current = segment;
}
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.SYLLABLE);
tw.setCurrentNode(current);
Element previous = (Element) tw.previousNode();
if (previous != null) {
assert previous.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE
+ ", got " + previous.getTagName();
}
return previous;
}
}
/**
* Retrieve the syllable two before the syllable belonging to this target.
*
* @author Marc Schröder
*
*/
public static class PrevPrevSyllableNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return null;
current = syllable;
} else { // boundary
current = segment;
}
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.SYLLABLE);
tw.setCurrentNode(current);
Element previous = (Element) tw.previousNode();
Element pp = (Element) tw.previousNode();
if (pp != null) {
assert pp.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE + ", got "
+ pp.getTagName();
}
return pp;
}
}
/**
* Retrieve the syllable following the syllable belonging to this target.
*
* @author Marc Schröder
*
*/
public static class NextSyllableNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return null;
current = syllable;
} else { // boundary
current = segment;
}
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.SYLLABLE);
tw.setCurrentNode(current);
Element next = (Element) tw.nextNode();
if (next != null) {
assert next.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE
+ ", got " + next.getTagName();
}
return next;
}
}
/**
* Retrieve the syllable two after the syllable belonging to this target.
*
* @author Marc Schröder
*
*/
public static class NextNextSyllableNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return null;
current = syllable;
} else { // boundary
current = segment;
}
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.SYLLABLE);
tw.setCurrentNode(current);
Element next = (Element) tw.nextNode();
Element nn = (Element) tw.nextNode();
if (nn != null) {
assert nn.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE + ", got "
+ nn.getTagName();
}
return nn;
}
}
/**
* Retrieve the word belonging to this target.
*
* @author Marc Schröder
*
*/
public static class WordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word != null) {
assert word.getTagName().equals(MaryXML.TOKEN) : "Unexpected tag name: expected " + MaryXML.TOKEN + ", got "
+ word.getTagName();
}
return word;
}
}
/** Last syllable in phrase. */
public static class LastSyllableInPhraseNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
Element last = (Element) tw.lastChild();
if (last != null) {
assert last.getTagName().equals(MaryXML.SYLLABLE) : "Unexpected tag name: expected " + MaryXML.SYLLABLE
+ ", got " + last.getTagName();
}
return last;
}
}
public static class NextWordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return null;
current = word;
} else { // boundary
current = segment;
}
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(current);
// The next word is the next token with a "ph" attribute:
Element nextWord = null;
Element nextToken;
while ((nextToken = (Element) tw.nextNode()) != null) {
if (nextToken.hasAttribute("ph")) {
nextWord = nextToken;
break;
}
}
if (nextWord != null) {
assert nextWord.getTagName().equals(MaryXML.TOKEN) : "Unexpected tag name: expected " + MaryXML.TOKEN + ", got "
+ nextWord.getTagName();
}
return nextWord;
}
}
public static class PrevWordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return null;
current = word;
} else { // boundary
current = segment;
}
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(current);
// The next word is the next token with a "ph" attribute:
Element prevWord = null;
Element prevToken;
while ((prevToken = (Element) tw.previousNode()) != null) {
if (prevToken.hasAttribute("ph")) {
prevWord = prevToken;
break;
}
}
if (prevWord != null) {
assert prevWord.getTagName().equals(MaryXML.TOKEN) : "Unexpected tag name: expected " + MaryXML.TOKEN + ", got "
+ prevWord.getTagName();
}
return prevWord;
}
}
public static class FirstSegmentNextWordNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return null;
current = word;
} else { // boundary
current = segment;
}
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(current);
// The next word is the next token with a "ph" attribute:
Element nextWord = null;
Element nextToken;
while ((nextToken = (Element) tw.nextNode()) != null) {
if (nextToken.hasAttribute("ph")) {
nextWord = nextToken;
break;
}
}
if (nextWord == null) {
return null;
}
assert nextWord.getTagName().equals(MaryXML.TOKEN) : "Unexpected tag name: expected " + MaryXML.TOKEN + ", got "
+ nextWord.getTagName();
TreeWalker sw = MaryDomUtils.createTreeWalker(nextWord, MaryXML.PHONE);
Element first = (Element) sw.firstChild();
if (first != null) {
assert first.getTagName().equals(MaryXML.PHONE) : "Unexpected tag name: expected " + MaryXML.PHONE + ", got "
+ first.getTagName();
}
return first;
}
}
public static class LastWordInSentenceNavigator implements TargetElementNavigator {
public Element getElement(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return null;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return null;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
Element lastWord = null;
Element lastToken = (Element) tw.lastChild();
// The last word is the lastToken which has a "ph" attribute:
while (lastToken != null) {
if (lastToken.hasAttribute("ph")) {
lastWord = lastToken;
break;
}
lastToken = (Element) tw.previousNode();
}
if (lastWord != null) {
assert lastWord.getTagName().equals(MaryXML.TOKEN) : "Unexpected tag name: expected " + MaryXML.TOKEN + ", got "
+ lastWord.getTagName();
}
return lastWord;
}
}
// no instances
protected MaryGenericFeatureProcessors() {
}
/**
* flite never returns an int more than 19 from a feature processor, we duplicate that behavior in the processors so that our
* tests will match. let's keep this number as a constant for better overview
*/
private static final int RAIL_LIMIT = 19;
private static final String[] ZERO_TO_NINETEEN = new String[] { "0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "10", "11",
"12", "13", "14", "15", "16", "17", "18", "19" };
/**
* Indicate whether a unit is an edge unit, which is never the case for a target.
*/
public static class Edge implements ByteValuedFeatureProcessor {
public String getName() {
return "edge";
}
public String[] getValues() {
return new String[] { "0", "start", "end" };
}
/**
* This processor always returns 0 for targets.
*/
public byte process(Target target) {
return (byte) 0;
}
}
/**
* Is the given half phone target a left or a right half?
*
* @author Marc Schröder
*
*/
public static class HalfPhoneLeftRight implements ByteValuedFeatureProcessor {
protected ByteStringTranslator values;
/**
* Initialise a HalfPhoneLeftRight feature processor.
*/
public HalfPhoneLeftRight() {
this.values = new ByteStringTranslator(new String[] { "0", "L", "R" });
}
public String getName() {
return "halfphone_lr";
}
public String[] getValues() {
return values.getStringValues();
}
public byte process(Target target) {
if (!(target instanceof HalfPhoneTarget))
return 0;
HalfPhoneTarget hpTarget = (HalfPhoneTarget) target;
String value = (hpTarget.isLeftHalf() ? "L" : "R");
return values.get(value);
}
}
/**
* Sentence Style for the given target
*
* @author Sathish Chandra Pammi
*
*/
public static class Style implements ByteValuedFeatureProcessor {
protected ByteStringTranslator values;
protected TargetElementNavigator navigator;
protected final String styleTagName = "style";
/**
* Initialize a speaking Style feature processor.
*/
public Style() {
this.values = new ByteStringTranslator(new String[] { "0", "neutral", "poker", "happy", "sad", "angry", "excited" });
this.navigator = new SegmentNavigator();
}
public String getName() {
return styleTagName;
}
public String[] getValues() {
return values.getStringValues();
}
public byte process(Target target) {
String style = null;
Element segment = target.getMaryxmlElement();
if (segment != null) {
Element prosody = (Element) MaryDomUtils.getClosestAncestorWithAttribute(segment, MaryXML.PROSODY, styleTagName);
if (prosody != null) {
style = prosody.getAttribute(styleTagName);
}
}
if (style == null || style.equals(""))
style = "0";
if (values.contains(style)) {
return values.get(style);
} else { // silently ignore unknown values
return 0;
}
}
}
/**
* Checks to see if the given syllable is accented.
*/
public static class Accented implements ByteValuedFeatureProcessor {
protected String name;
protected TargetElementNavigator navigator;
public Accented(String name, TargetElementNavigator syllableNavigator) {
this.name = name;
this.navigator = syllableNavigator;
}
public String getName() {
return name;
}
public String[] getValues() {
return new String[] { "0", "1" };
}
/**
* Performs some processing on the given item.
*
* @param target
* the target to process
* @return "1" if the syllable is accented; otherwise "0"
*/
public byte process(Target target) {
Element syllable = navigator.getElement(target);
if (syllable != null && syllable.hasAttribute("accent")) {
return (byte) 1;
} else {
return (byte) 0;
}
}
}
/**
* Checks to see if the given syllable is stressed.
*/
public static class Stressed implements ByteValuedFeatureProcessor {
protected String name;
protected TargetElementNavigator navigator;
public Stressed(String name, TargetElementNavigator syllableNavigator) {
this.name = name;
this.navigator = syllableNavigator;
}
public String getName() {
return name;
}
public String[] getValues() {
return new String[] { "0", "1" };
}
/**
* Performs some processing on the given item.
*
* @param target
* the target to process
* @return "1" if the syllable is stressed; otherwise "0"
*/
public byte process(Target target) {
Element syllable = navigator.getElement(target);
if (syllable == null)
return 0;
String value = syllable.getAttribute("stress");
if (value.equals(""))
return 0;
byte stressValue = Byte.parseByte(value);
if (stressValue > 1) {
// out of range, set to 1
stressValue = 1;
}
return stressValue;
}
}
/**
* Syllable tone for the given target
*
* @author sathish pammi
*
*/
public static class SyllableTone implements ByteValuedFeatureProcessor {
protected String name;
protected TargetElementNavigator navigator;
public SyllableTone(String name, TargetElementNavigator syllableNavigator) {
this.name = name;
this.navigator = syllableNavigator;
}
public String getName() {
return name;
}
public String[] getValues() {
return new String[] { "0", "1", "2", "3", "4" };
}
/**
* Performs some processing on the given item.
*
* @param target
* the target to process
* @return tone value
*/
public byte process(Target target) {
Element syllable = navigator.getElement(target);
if (syllable == null)
return 0;
String value = syllable.getAttribute("tone");
if (value.equals(""))
return 0;
byte toneValue = Byte.parseByte(value);
if (toneValue > 4 || toneValue < 1) {
// out of range, set to 0
toneValue = 0;
}
return toneValue;
}
}
/**
* Returns as a byte the number of phrases in the current sentence.
*/
public static class SentenceNumPhrases implements ByteValuedFeatureProcessor {
public SentenceNumPhrases() {
}
public String getName() {
return "sentence_numphrases";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
*
* @return the number of phrases in the sentence
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.PHRASE);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Returns as an Integer the number of words in the current sentence. This is a feature processor. A feature processor takes
* an item, performs some sort of processing on the item and returns an object.
*/
public static class SentenceNumWords implements ByteValuedFeatureProcessor {
public SentenceNumWords() {
}
public String getName() {
return "sentence_numwords";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @return the number of words in the sentence
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
// only tokens with a "ph" attribute count as words:
if (e.hasAttribute("ph"))
count++;
}
return (byte) count;
}
}
/**
* Returns as a byte the number of phrases in the current sentence.
*/
public static class PhraseNumSyls implements ByteValuedFeatureProcessor {
public PhraseNumSyls() {
}
public String getName() {
return "phrase_numsyls";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @return the number of words in the phrase
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Returns as a byte the number of words in the current phrase.
*/
public static class PhraseNumWords implements ByteValuedFeatureProcessor {
public PhraseNumWords() {
}
public String getName() {
return "phrase_numwords";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* Performs some processing on the given item.
*
* @param target
* the item to process
*
* @return the number of words in the phrase
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.TOKEN);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Returns as an Integer the number of syllables in the given word. This is a feature processor. A feature processor takes an
* item, performs some sort of processing on the item and returns an object.
*/
public static class WordNumSyls implements ByteValuedFeatureProcessor {
public WordNumSyls() {
}
public String getName() {
return "word_numsyls";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @return the number of syllables in the given word
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.SYLLABLE);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Returns as a byte the number of segments in the given word.
*/
public static class WordNumSegs implements ByteValuedFeatureProcessor {
public WordNumSegs() {
}
public String getName() {
return "word_numsegs";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @return the number of segments in the given word
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.PHONE);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Returns as an Integer the number of segments in the current syllable. This is a feature processor. A feature processor
* takes an item, performs some sort of processing on the item and returns an object.
*/
public static class SylNumSegs implements ByteValuedFeatureProcessor {
public SylNumSegs() {
}
public String getName() {
return "syl_numsegs";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* Performs some processing on the given item.
*
* @param target
* the item to process
*
* @return the number of segments in the given word
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
if (!segment.getTagName().equals(MaryXML.PHONE))
return 0;
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(syllable, MaryXML.PHONE);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* @deprecated use SegsFromSylStart instead
*/
public static class PosInSyl extends SegsFromSylStart {
public PosInSyl() {
super();
}
public String getName() {
return "pos_in_syl";
}
}
/**
* Finds the position of the phone in the syllable.
*/
public static class SegsFromSylStart implements ByteValuedFeatureProcessor {
public SegsFromSylStart() {
}
public String getName() {
return "segs_from_syl_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the position of the phone in the syllable
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
if (!segment.getTagName().equals(MaryXML.PHONE))
return 0;
int count = 0;
Element e = segment;
while ((e = MaryDomUtils.getPreviousSiblingElement(e)) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Finds the position of the phone from the end of the syllable.
*/
public static class SegsFromSylEnd implements ByteValuedFeatureProcessor {
public SegsFromSylEnd() {
}
public String getName() {
return "segs_from_syl_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the position of the phone in the syllable
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
if (!segment.getTagName().equals(MaryXML.PHONE))
return 0;
int count = 0;
Element e = segment;
while ((e = MaryDomUtils.getNextSiblingElement(e)) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Finds the position of the segment from the start of the word.
*/
public static class SegsFromWordStart implements ByteValuedFeatureProcessor {
public SegsFromWordStart() {
}
public String getName() {
return "segs_from_word_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the position of the phone in the syllable
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.PHONE);
tw.setCurrentNode(segment);
int count = 0;
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Finds the position of the segment from the end of the word.
*/
public static class SegsFromWordEnd implements ByteValuedFeatureProcessor {
public SegsFromWordEnd() {
}
public String getName() {
return "segs_from_word_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* Performs some processing on the given item.
*
* @param target
* the target to process
* @return the position of the phone in the syllable
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return (byte) 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(word, MaryXML.PHONE);
tw.setCurrentNode(segment);
int count = 0;
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Finds the position of the syllable from the start of the word.
*/
public static class SylsFromWordStart implements ByteValuedFeatureProcessor {
public SylsFromWordStart() {
}
public String getName() {
return "syls_from_word_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the position of the syllable in the word
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
if (!segment.getTagName().equals(MaryXML.PHONE))
return 0;
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return (byte) 0;
int count = 0;
Element e = syllable;
while ((e = MaryDomUtils.getPreviousSiblingElement(e)) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Finds the position of the syllable from the end of the word.
*/
public static class SylsFromWordEnd implements ByteValuedFeatureProcessor {
public SylsFromWordEnd() {
}
public String getName() {
return "syls_from_word_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the position of the phone in the syllable
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return (byte) 0;
if (!segment.getTagName().equals(MaryXML.PHONE))
return 0;
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return (byte) 0;
int count = 0;
Element e = syllable;
while ((e = MaryDomUtils.getNextSiblingElement(e)) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Determines the break level after this syllable.
*/
public static class SylBreak implements ByteValuedFeatureProcessor {
protected String name;
protected TargetElementNavigator navigator;
public SylBreak(String name, TargetElementNavigator syllableNavigator) {
this.name = name;
this.navigator = syllableNavigator;
}
public String getName() {
return name;
}
/**
* "4" for a big break, "3" for a break; "1" = word-final; "0" = within-word
*/
public String[] getValues() {
return new String[] { "0", "1", "unused", "3", "4" };
}
/**
* @param target
* the target to process
* @return the break level after the syllable returned by syllableNavigator
*/
public byte process(Target target) {
Element syllable = navigator.getElement(target);
if (syllable == null)
return 0;
// is there another syllable following in the token?
if (MaryDomUtils.getNextSiblingElement(syllable) != null)
return 0;
// else, it is at least word-final.
Element word = (Element) syllable.getParentNode();
if (word == null)
return 0;
assert word.getTagName().equals(MaryXML.TOKEN) : "Unexpected tag name: expected " + MaryXML.TOKEN + ", got "
+ word.getTagName();
Element sentence = (Element) MaryDomUtils.getAncestor(word, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN, MaryXML.BOUNDARY);
tw.setCurrentNode(word);
// The next word is the next token with a "ph" attribute:
Element e;
while ((e = (Element) tw.nextNode()) != null) {
if (e.getTagName().equals(MaryXML.BOUNDARY) || e.getTagName().equals(MaryXML.TOKEN) && e.hasAttribute("ph"))
break;
}
if (e == null) {
// we are the last token in the sentence, but there is no boundary...
// OK, let's say it is sentence-final anyway:
return 4;
}
if (e.getTagName().equals(MaryXML.TOKEN)) {
// a word follows
return 1;
}
assert e.getTagName().equals(MaryXML.BOUNDARY) : "Unexpected tag name: expected " + MaryXML.BOUNDARY + ", got "
+ e.getTagName();
String bi = e.getAttribute("breakindex");
if (bi.equals("")) {
// no breakindex
return 1;
}
try {
int ibi = Integer.parseInt(bi);
if (ibi >= 4)
return 4;
if (ibi == 3)
return 3;
} catch (NumberFormatException nfe) {
}
// default: a word boundary
return 1;
}
}
/**
* Classifies the the syllable as single, initial, mid or final.
*/
public static class PositionType implements ByteValuedFeatureProcessor {
protected TargetElementNavigator navigator;
protected ByteStringTranslator values;
public PositionType() {
values = new ByteStringTranslator(new String[] { "0", "single", "final", "initial", "mid" });
navigator = new SyllableNavigator();
}
public String getName() {
return "position_type";
}
public String[] getValues() {
return values.getStringValues();
}
/**
* @param target
* the target to process
* @return classifies the syllable as "single", "final", "initial" or "mid"
*/
public byte process(Target target) {
Element syllable = navigator.getElement(target);
if (syllable == null)
return 0;
String type;
if (MaryDomUtils.getNextSiblingElement(syllable) == null) {
if (MaryDomUtils.getPreviousSiblingElement(syllable) == null) {
type = "single";
} else {
type = "final";
}
} else if (MaryDomUtils.getPreviousSiblingElement(syllable) == null) {
type = "initial";
} else {
type = "mid";
}
return values.get(type);
}
}
/**
* Checks if segment is a pause.
*/
public static class IsPause implements ByteValuedFeatureProcessor {
protected TargetElementNavigator navigator;
protected String name;
public IsPause(String name, TargetElementNavigator segmentNavigator) {
this.name = name;
this.navigator = segmentNavigator;
}
public String getName() {
return name;
}
public String[] getValues() {
return new String[] { "0", "1" };
}
/**
* Check if segment is a pause
*
* @param target
* the target to process
* @return 0 if false, 1 if true
*/
public byte process(Target target) {
Element seg = navigator.getElement(target);
if (seg == null)
return 0;
if (seg.getTagName().equals(MaryXML.BOUNDARY))
return 1;
return 0;
}
}
public static class BreakIndex implements ByteValuedFeatureProcessor {
protected ByteStringTranslator values;
public BreakIndex() {
values = new ByteStringTranslator(new String[] { "0", "1", "2", "3", "4", "5", "6" });
}
public String getName() {
return "breakindex";
}
public String[] getValues() {
return values.getStringValues();
}
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word == null)
return 0;
// is there another segment following in the token?
TreeWalker tww = MaryDomUtils.createTreeWalker(word, MaryXML.PHONE);
tww.setCurrentNode(segment);
if (tww.nextNode() != null)
return 0;
// else, it is at least word-final.
Element sentence = (Element) MaryDomUtils.getAncestor(word, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN, MaryXML.BOUNDARY);
tw.setCurrentNode(word);
// The next word is the next token with a "ph" attribute:
Element e;
while ((e = (Element) tw.nextNode()) != null) {
if (e.getTagName().equals(MaryXML.BOUNDARY) || e.getTagName().equals(MaryXML.TOKEN) && e.hasAttribute("ph"))
break;
}
if (e == null) {
// we are the last token in the sentence, but there is no boundary...
// OK, let's say it is sentence-final anyway:
return 4;
}
if (e.getTagName().equals(MaryXML.TOKEN)) {
// a word follows
return 1;
}
assert e.getTagName().equals(MaryXML.BOUNDARY) : "Unexpected tag name: expected " + MaryXML.BOUNDARY + ", got "
+ e.getTagName();
String bi = e.getAttribute("breakindex");
if (bi.equals("")) {
// no breakindex
return 1;
}
try {
int ibi = Integer.parseInt(bi);
if (ibi > 6)
ibi = 6;
if (ibi < 2)
ibi = 2;
return (byte) ibi;
} catch (NumberFormatException nfe) {
}
// default: a word boundary
return 1;
}
}
/**
* The ToBI accent of the current syllable.
*/
public static class TobiAccent implements ByteValuedFeatureProcessor {
protected String name;
protected TargetElementNavigator navigator;
protected ByteStringTranslator values;
public TobiAccent(String name, TargetElementNavigator syllableNavigator) {
this.name = name;
this.navigator = syllableNavigator;
this.values = new ByteStringTranslator(new String[] { "0", "*", "H*", "!H*", "^H*", "L*", "L+H*", "L*+H", "L+!H*",
"L*+!H", "L+^H*", "L*+^H", "H+L*", "H+!H*", "H+^H*", "!H+!H*", "^H+!H*", "^H+^H*", "H*+L", "!H*+L" });
}
public String getName() {
return name;
}
public String[] getValues() {
return values.getStringValues();
}
/**
* For the given syllable item, return its tobi accent, or 0 if there is none.
*/
public byte process(Target target) {
Element syllable = navigator.getElement(target);
if (syllable == null)
return 0;
String accent = syllable.getAttribute("accent");
if (accent.equals("")) {
return 0;
}
return values.get(accent);
}
}
/**
* The ToBI endtone associated with the current syllable.
*/
public static class TobiEndtone implements ByteValuedFeatureProcessor {
protected String name;
protected TargetElementNavigator navigator;
protected ByteStringTranslator values;
public TobiEndtone(String name, TargetElementNavigator syllableNavigator) {
this.name = name;
this.navigator = syllableNavigator;
this.values = new ByteStringTranslator(new String[] { "0", "H-", "!H-", "L-", "H-%", "!H-%", "H-^H%", "!H-^H%",
"L-H%", "L-%", "L-L%", "H-H%", "H-L%" });
}
public String getName() {
return name;
}
public String[] getValues() {
return values.getStringValues();
}
/**
* For the given syllable item, return its tobi end tone, or 0 if there is none.
*/
public byte process(Target target) {
Element syllable = navigator.getElement(target);
if (syllable == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(syllable, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.SYLLABLE, MaryXML.BOUNDARY);
tw.setCurrentNode(syllable);
Element e = (Element) tw.nextNode();
if (e == null)
return 0;
if (e.getTagName().equals(MaryXML.SYLLABLE))
return 0;
assert e.getTagName().equals(MaryXML.BOUNDARY) : "Unexpected tag name: expected " + MaryXML.BOUNDARY + ", got "
+ e.getTagName();
String endtone = e.getAttribute("tone");
if (endtone.equals("")) {
return 0;
}
return values.get(endtone);
}
}
/**
* The next ToBI accent following the current syllable in the current phrase.
*/
public static class NextAccent extends TobiAccent {
public NextAccent() {
super("next_accent", null);
}
/**
* Search for an accented syllable, and return its tobi accent, or 0 if there is none.
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return 0;
current = syllable;
} else { // boundary
current = segment;
}
Element phrase = (Element) MaryDomUtils.getAncestor(current, MaryXML.PHRASE);
if (phrase == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
tw.setCurrentNode(current);
Element s;
while ((s = (Element) tw.nextNode()) != null) {
if (s.hasAttribute("accent")) {
String accent = s.getAttribute("accent");
return values.get(accent);
}
}
return 0;
}
}
/**
* The previous ToBI accent preceding the current syllable in the current phrase.
*/
public static class PrevAccent extends TobiAccent {
public PrevAccent() {
super("prev_accent", null);
}
/**
* Search for an accented syllable, and return its tobi accent, or 0 if there is none.
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element current;
if (segment.getTagName().equals(MaryXML.PHONE)) {
Element syllable = (Element) segment.getParentNode();
if (syllable == null)
return 0;
current = syllable;
} else { // boundary
current = segment;
}
Element phrase = (Element) MaryDomUtils.getAncestor(current, MaryXML.PHRASE);
if (phrase == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
tw.setCurrentNode(current);
Element s;
while ((s = (Element) tw.previousNode()) != null) {
if (s.hasAttribute("accent")) {
String accent = s.getAttribute("accent");
return values.get(accent);
}
}
return 0;
}
}
/**
* The ToBI endtone associated with the last syllable of the current phrase.
*/
public static class PhraseEndtone extends TobiEndtone {
public PhraseEndtone() {
super("phrase_endtone", new LastSyllableInPhraseNavigator());
}
}
/**
* The ToBI endtone associated with the last syllable of the previous phrase.
*/
public static class PrevPhraseEndtone extends TobiEndtone {
public PrevPhraseEndtone() {
super("prev_phrase_endtone", null);
}
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
Document doc = phrase.getOwnerDocument();
TreeWalker tw = MaryDomUtils.createTreeWalker(doc, doc, MaryXML.BOUNDARY);
tw.setCurrentNode(phrase);
Element boundary = (Element) tw.previousNode();
if (boundary == null)
return 0;
String endtone = boundary.getAttribute("tone");
if (endtone.equals("")) {
return 0;
}
return values.get(endtone);
}
}
/**
* Counts the number of syllables since the start of the phrase.
*/
public static class SylsFromPhraseStart implements ByteValuedFeatureProcessor {
public SylsFromPhraseStart() {
}
public String getName() {
return "syls_from_phrase_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of syllables since the last major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
Element syllable = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SYLLABLE);
if (syllable != null) {
tw.setCurrentNode(syllable);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Counts the number of syllables until the end of the phrase.
*/
public static class SylsFromPhraseEnd implements ByteValuedFeatureProcessor {
public SylsFromPhraseEnd() {
}
public String getName() {
return "syls_from_phrase_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of accented syllables since the last major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Counts the number of stressed syllables since the start of the phrase.
*/
public static class StressedSylsFromPhraseStart implements ByteValuedFeatureProcessor {
public StressedSylsFromPhraseStart() {
}
public String getName() {
return "stressed_syls_from_phrase_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of stressed syllables since the last major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
Element syllable = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SYLLABLE);
if (syllable != null) {
tw.setCurrentNode(syllable);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
String stress = e.getAttribute("stress");
if (stress.equals("1"))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of stressed syllables until the end of the phrase.
*/
public static class StressedSylsFromPhraseEnd implements ByteValuedFeatureProcessor {
public StressedSylsFromPhraseEnd() {
}
public String getName() {
return "stressed_syls_from_phrase_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of stressed syllables since the last major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
String stress = e.getAttribute("stress");
if (stress.equals("1"))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of accented syllables since the start of the phrase.
*/
public static class AccentedSylsFromPhraseStart implements ByteValuedFeatureProcessor {
public AccentedSylsFromPhraseStart() {
}
public String getName() {
return "accented_syls_from_phrase_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of accented syllables since the last major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
Element syllable = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SYLLABLE);
if (syllable != null) {
tw.setCurrentNode(syllable);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
String accent = e.getAttribute("accent");
if (!accent.equals(""))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of accented syllables until the end of the phrase.
*/
public static class AccentedSylsFromPhraseEnd implements ByteValuedFeatureProcessor {
public AccentedSylsFromPhraseEnd() {
}
public String getName() {
return "accented_syls_from_phrase_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of accented syllables since the last major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
String accent = e.getAttribute("accent");
if (!accent.equals(""))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of words since the start of the phrase.
*/
public static class WordsFromPhraseStart implements ByteValuedFeatureProcessor {
public WordsFromPhraseStart() {
}
public String getName() {
return "words_from_phrase_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of words since the last major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.TOKEN);
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word != null) {
tw.setCurrentNode(word);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
// only count tokens that have a "ph" attribute:
if (e.hasAttribute("ph"))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of words until the end of the phrase.
*/
public static class WordsFromPhraseEnd implements ByteValuedFeatureProcessor {
public WordsFromPhraseEnd() {
}
public String getName() {
return "words_from_phrase_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of words until the next major break
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.TOKEN);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
// only count tokens that have a "ph" attribute
if (e.hasAttribute("ph"))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of words since the start of the sentence.
*/
public static class WordsFromSentenceStart implements ByteValuedFeatureProcessor {
public WordsFromSentenceStart() {
}
public String getName() {
return "words_from_sentence_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of words since the beginning of the sentence
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
Element word = (Element) MaryDomUtils.getAncestor(segment, MaryXML.TOKEN);
if (word != null) {
tw.setCurrentNode(word);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
// only count tokens that have a "ph" attribute:
if (e.hasAttribute("ph"))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of words until the end of the sentence.
*/
public static class WordsFromSentenceEnd implements ByteValuedFeatureProcessor {
public WordsFromSentenceEnd() {
}
public String getName() {
return "words_from_sentence_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of words until the end of the sentence
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
// only count tokens that have a "ph" attribute:
if (e.hasAttribute("ph"))
count++;
}
return (byte) count;
}
}
/**
* Counts the number of phrases since the start of the sentence.
*/
public static class PhrasesFromSentenceStart implements ByteValuedFeatureProcessor {
public PhrasesFromSentenceStart() {
}
public String getName() {
return "phrases_from_sentence_start";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of phrases since the start of the sentence
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.PHRASE);
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase != null) {
tw.setCurrentNode(phrase);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Counts the number of phrases until the end of the sentence.
*/
public static class PhrasesFromSentenceEnd implements ByteValuedFeatureProcessor {
public PhrasesFromSentenceEnd() {
}
public String getName() {
return "phrases_from_sentence_end";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of phrases until the end of the sentence.
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SENTENCE);
if (sentence == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.PHRASE);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
}
return (byte) count;
}
}
/**
* Counts the number of syllables since the last accent in the current phrase.
*/
public static class SylsFromPrevAccent implements ByteValuedFeatureProcessor {
public SylsFromPrevAccent() {
}
public String getName() {
return "syls_from_prev_accent";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of syllables since the last accent
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
Element syllable = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SYLLABLE);
if (syllable != null) {
tw.setCurrentNode(syllable);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
count++;
String accent = e.getAttribute("accent");
if (!accent.equals(""))
break;
}
return (byte) count;
}
}
/**
* Counts the number of syllables until the next accent in the current phrase.
*/
public static class SylsToNextAccent implements ByteValuedFeatureProcessor {
public SylsToNextAccent() {
}
public String getName() {
return "syls_to_next_accent";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of syllables until the next accent
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
String accent = e.getAttribute("accent");
if (!accent.equals(""))
break;
}
return (byte) count;
}
}
/**
* Counts the number of syllables since the last stressed syllable in the current phrase.
*/
public static class SylsFromPrevStressed implements ByteValuedFeatureProcessor {
public SylsFromPrevStressed() {
}
public String getName() {
return "syls_from_prev_stressed";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of syllables since the last accent
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
Element syllable = (Element) MaryDomUtils.getAncestor(segment, MaryXML.SYLLABLE);
if (syllable != null) {
tw.setCurrentNode(syllable);
} else {
tw.setCurrentNode(segment);
}
Element e;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
count++;
String stress = e.getAttribute("stress");
if (stress.equals("1"))
break;
}
return (byte) count;
}
}
/**
* Counts the number of syllables until the next stressed syllable in the current phrase.
*/
public static class SylsToNextStressed implements ByteValuedFeatureProcessor {
public SylsToNextStressed() {
}
public String getName() {
return "syls_to_next_stressed";
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
/**
* @param target
* the target to process
* @return the number of syllables until the next stressed syllable
*/
public byte process(Target target) {
Element segment = target.getMaryxmlElement();
if (segment == null)
return 0;
Element phrase = (Element) MaryDomUtils.getAncestor(segment, MaryXML.PHRASE);
if (phrase == null)
return 0;
int count = 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(phrase, MaryXML.SYLLABLE);
tw.setCurrentNode(segment);
Element e;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
String stress = e.getAttribute("stress");
if (stress.equals("1"))
break;
}
return (byte) count;
}
}
/**
* Determines the word punctuation.
*/
public static class WordPunc implements ByteValuedFeatureProcessor {
protected String name;
protected TargetElementNavigator navigator;
protected ByteStringTranslator values;
/**
* @param name
* name of this feature processor
* @param wordNavigator
* a navigator which returns a word for a target. This navigator decides the word for which the punctuation
* will be computed.
*/
public WordPunc(String name, TargetElementNavigator wordNavigator) {
this.name = name;
this.navigator = wordNavigator;
this.values = new ByteStringTranslator(new String[] { "0", ".", ",", ";", ":", "(", ")", "?", "!", "\"" });
}
public String getName() {
return name;
}
public String[] getValues() {
return values.getStringValues();
}
public byte process(Target target) {
Element word = navigator.getElement(target);
if (word == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(word, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN, MaryXML.BOUNDARY);
tw.setCurrentNode(word);
Element next = (Element) tw.nextNode();
if (next == null || !next.getTagName().equals(MaryXML.TOKEN) || next.hasAttribute("ph"))
return 0;
String text = MaryDomUtils.tokenText(next);
if (values.contains(text)) {
return values.get(text);
}
// unknown or no punctuation: return "0"
return values.get("0");
}
}
/**
* Determines the next word punctuation in the sentence.
*/
public static class NextPunctuation extends WordPunc {
public NextPunctuation() {
super("next_punctuation", new WordNavigator());
}
public byte process(Target target) {
Element word = navigator.getElement(target);
if (word == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(word, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(word);
Element e;
while ((e = (Element) tw.nextNode()) != null) {
if (e.hasAttribute("ph")) // a word
continue;
// potentially a punctuation
String text = MaryDomUtils.tokenText(e);
if (values.contains(text)) {
return values.get(text);
}
}
// no next punctuation: return "0"
return values.get("0");
}
}
/**
* Determines the previous word punctuation in the sentence.
*/
public static class PrevPunctuation extends WordPunc {
public PrevPunctuation() {
super("prev_punctuation", new WordNavigator());
}
public byte process(Target target) {
Element word = navigator.getElement(target);
if (word == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(word, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(word);
Element e;
while ((e = (Element) tw.previousNode()) != null) {
if (e.hasAttribute("ph")) // a word
continue;
// potentially a punctuation
String text = MaryDomUtils.tokenText(e);
if (values.contains(text)) {
return values.get(text);
}
}
// no next punctuation: return "0"
return values.get("0");
}
}
/**
* Determines the distance in words to the next word punctuation in the sentence.
*/
public static class WordsToNextPunctuation extends WordPunc {
public WordsToNextPunctuation() {
super("words_to_next_punctuation", new WordNavigator());
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
public byte process(Target target) {
Element word = navigator.getElement(target);
if (word == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(word, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(word);
Element e;
int count = 0;
while ((e = (Element) tw.nextNode()) != null && count < RAIL_LIMIT) {
count++;
if (e.hasAttribute("ph")) // a word
continue;
// potentially a punctuation
String text = MaryDomUtils.tokenText(e);
if (values.contains(text)) {
break;
}
}
// found punctuation or end of sentence:
return (byte) count;
}
}
/**
* Determines the distance in words from the previous word punctuation in the sentence.
*/
public static class WordsFromPrevPunctuation extends WordPunc {
public WordsFromPrevPunctuation() {
super("words_from_prev_punctuation", new WordNavigator());
}
public String[] getValues() {
return ZERO_TO_NINETEEN;
}
public byte process(Target target) {
Element word = navigator.getElement(target);
if (word == null)
return 0;
Element sentence = (Element) MaryDomUtils.getAncestor(word, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.TOKEN);
tw.setCurrentNode(word);
Element e;
int count = 0;
while ((e = (Element) tw.previousNode()) != null && count < RAIL_LIMIT) {
count++;
if (e.hasAttribute("ph")) // a word
continue;
// potentially a punctuation
String text = MaryDomUtils.tokenText(e);
if (values.contains(text)) {
break;
}
}
// found punctuation or start of sentence:
return (byte) count;
}
}
/**
* Determine the prosodic property of a target
*
* @author Anna Hunecke
*
*/
public static class Selection_Prosody implements ByteValuedFeatureProcessor {
protected TargetElementNavigator navigator;
protected ByteStringTranslator values = new ByteStringTranslator(new String[] { "0", "stressed", "pre-nuclear",
"nuclear", "finalHigh", "finalLow", "final" });
private Set<String> lowEndtones = new HashSet<String>(Arrays.asList(new String[] { "L-", "L-%", "L-L%" }));
private Set<String> highEndtones = new HashSet<String>(Arrays.asList(new String[] { "H-", "!H-", "H-%", "H-L%", "!H-%",
"H-^H%", "!H-^H%", "L-H%", "H-H%" }));
public Selection_Prosody(TargetElementNavigator syllableNavigator) {
this.navigator = syllableNavigator;
}
public String getName() {
return "selection_prosody";
}
public String[] getValues() {
return values.getStringValues();
}
/**
* Determine the prosodic property of the target
*
* @param target
* the target
* @return 0 - unstressed, 1 - stressed, 2 - pre-nuclear accent 3 - nuclear accent, 4 - phrase final high, 5 - phrase
* final low, 6 - phrase final (with unknown high/low status).
*/
public byte process(Target target) {
// first find out if syllable is stressed
Element syllable = navigator.getElement(target);
if (syllable == null)
return (byte) 0;
boolean stressed = false;
if (syllable.getAttribute("stress").equals("1")) {
stressed = true;
}
// find out if we have an accent
boolean accented = syllable.hasAttribute("accent");
boolean nuclear = true; // relevant only if accented == true
// find out the position of the target
boolean phraseFinal = false;
String endtone = null;
Element sentence = (Element) MaryDomUtils.getAncestor(syllable, MaryXML.SENTENCE);
if (sentence == null)
return 0;
TreeWalker tw = MaryDomUtils.createTreeWalker(sentence, MaryXML.SYLLABLE, MaryXML.BOUNDARY);
tw.setCurrentNode(syllable);
Element e = (Element) tw.nextNode();
if (e != null) {
if (e.getTagName().equals(MaryXML.BOUNDARY)) {
phraseFinal = true;
endtone = e.getAttribute("tone");
}
if (accented) { // look forward for any accent
while (e != null) {
if (e.getTagName().equals(MaryXML.SYLLABLE) && e.hasAttribute("accent")) {
nuclear = false;
break;
}
e = (Element) tw.nextNode();
}
}
}
// Now, we know:
// stressed or not
// accented or not
// if accented, nuclear or not
// if final, the endtone
if (accented) {
if (nuclear) {
return values.get("nuclear");
} else {
return values.get("pre-nuclear");
}
} else if (phraseFinal) {
if (endtone != null && highEndtones.contains(endtone)) {
return values.get("finalHigh");
} else if (endtone != null && lowEndtones.contains(endtone)) {
return values.get("finalLow");
} else {
return values.get("final");
}
} else if (stressed) {
return values.get("stressed");
}
return (byte) 0;// return unstressed
}
}
/**
* Returns the duration of the given segment, in seconds.
*/
public static class UnitDuration implements ContinuousFeatureProcessor {
public String getName() {
return "unit_duration";
}
public float process(Target target) {
if (target instanceof DiphoneTarget) {
DiphoneTarget diphone = (DiphoneTarget) target;
return process(diphone.left) + process(diphone.right);
}
Element seg = target.getMaryxmlElement();
if (seg == null) {
return 0;
}
float phoneDuration = 0;
String sDur;
if (seg.getTagName().equals(MaryXML.PHONE))
sDur = seg.getAttribute("d");
else {
assert seg.getTagName().equals(MaryXML.BOUNDARY) : "segment should be a phone or a boundary, but is a "
+ seg.getTagName();
sDur = seg.getAttribute("duration");
}
if (sDur.equals("")) {
return 0;
}
try {
// parse duration string, and convert from milliseconds into seconds:
phoneDuration = Float.parseFloat(sDur) * 0.001f;
} catch (NumberFormatException nfe) {
}
if (target instanceof HalfPhoneTarget)
return phoneDuration / 2;
return phoneDuration;
}
}
/**
* Calculates the log of the fundamental frequency in the middle of a unit segment. This processor should be used by target
* items only -- for unit features during voice building, the actual measured values should be used.
*/
public static class UnitLogF0 implements ContinuousFeatureProcessor {
public String getName() {
return "unit_logf0";
}
public float process(Target target) {
return process(target, false);
}
/**
* Compute log f0 and log f0 delta for the given target.
*
* @param target
* target
* @param delta
* if true, return the delta, i.e. the logF0 slope; if false, return the log f0 value itself.
* @return 0 if seg is null, or if !seg.getTagName().equals(MaryXML.PHONE), 0 if lastPos == null or nextPos == null,
* return lastF0 otherwise; if delta return slope, return f0 otherwise
*/
protected float process(Target target, boolean delta) {
// Note: all variables in this method with "f0" in their name
// actually represent log f0 values.
if (target instanceof DiphoneTarget) {
DiphoneTarget diphone = (DiphoneTarget) target;
return (process(diphone.left) + process(diphone.right)) / 2;
}
// Idea: find the closest f0 targets in the current syllable, left and right of our middle;
// linearly interpolate between them to find the value in the middle of this unit.
Element seg = target.getMaryxmlElement();
if (seg == null) {
return 0;
}
if (!seg.getTagName().equals(MaryXML.PHONE)) {
return 0;
}
// get mid position of segment wrt phone start (phone start = 0, phone end = phone duration)
float mid;
float phoneDuration = getDuration(seg);
if (target instanceof HalfPhoneTarget) {
if (((HalfPhoneTarget) target).isLeftHalf()) {
mid = .25f;
} else {
mid = .75f;
}
} else { // phone target
mid = .5f;
}
// Now mid is the middle of the unit relative to the phone start, in percent
Float lastPos = null; // position relative to mid, in milliseconds (negative)
float lastF0 = 0;
Float nextPos = null; // position relative to mid, in milliseconds
float nextF0 = 0;
Float[] f0values = getLogF0Values(seg);
assert f0values != null;
// values are position, f0, position, f0, etc.;
// position is in percent of phone duration between 0 and 1, f0 is in Hz
for (int i = 0; i < f0values.length; i += 2) {
float pos = f0values[i];
if (pos <= mid) {
lastPos = (pos - mid) * phoneDuration; // negative or zero
lastF0 = f0values[i + 1];
} else if (pos > mid) {
nextPos = (pos - mid) * phoneDuration; // positive
nextF0 = f0values[i + 1];
break; // no point looking further to the right
}
}
if (lastPos == null) { // need to look to the left
float msBack = -mid * phoneDuration;
Element e = seg;
// get all phone units in the same phrase
Element phraseElement = (Element) MaryDomUtils.getAncestor(seg, MaryXML.PHRASE);
TreeWalker tw = MaryDomUtils.createTreeWalker(seg.getOwnerDocument(), phraseElement, MaryXML.PHONE);
Element en;
while ((en = (Element) tw.nextNode()) != null) {
if (en == seg) {
break;
}
}
while ((e = (Element) tw.previousNode()) != null) {
float dur = getDuration(e);
f0values = getLogF0Values(e);
if (f0values.length == 0) {
msBack -= dur;
continue;
}
assert f0values.length > 1;
float pos = f0values[f0values.length - 2];
lastPos = msBack - (1 - pos) * dur;
lastF0 = f0values[f0values.length - 1];
break;
}
}
if (nextPos == null) { // need to look to the right
float msForward = (1 - mid) * phoneDuration;
Element e = seg;
// get all phone units in the same phrase
Element phraseElement = (Element) MaryDomUtils.getAncestor(seg, MaryXML.PHRASE);
TreeWalker tw = MaryDomUtils.createTreeWalker(seg.getOwnerDocument(), phraseElement, MaryXML.PHONE);
Element en;
while ((en = (Element) tw.nextNode()) != null) {
if (en == seg) {
break;
}
}
while ((e = (Element) tw.nextNode()) != null) {
float dur = getDuration(e);
f0values = getLogF0Values(e);
if (f0values.length == 0) {
msForward += dur;
continue;
}
assert f0values.length > 1;
float pos = f0values[0];
nextPos = msForward + pos * dur;
nextF0 = f0values[1];
break;
}
}
if (lastPos == null && nextPos == null) {
// no info
return 0;
} else if (lastPos == null) {
// have only nextF0;
if (delta)
return 0;
else
return nextF0;
} else if (nextPos == null) {
// have only lastF0
if (delta)
return 0;
else
return lastF0;
}
assert lastPos <= 0 && 0 <= nextPos : "unexpected: lastPos=" + lastPos + ", nextPos=" + nextPos;
// build a linear function (f(x) = slope*x+intersectionYAxis)
float f0;
float slope;
if (lastPos - nextPos == 0) {
f0 = (lastF0 + nextF0) / 2;
slope = 0;
} else {
slope = (nextF0 - lastF0) / (nextPos - lastPos);
// calculate the pitch
f0 = lastF0 + slope * (-lastPos);
}
assert !Float.isNaN(f0) : "f0 is not a number";
assert lastF0 <= f0 && nextF0 >= f0 || lastF0 >= f0 && nextF0 <= f0 : "f0 should be between last and next values";
if (delta)
return slope;
else
return f0;
}
private Float[] getLogF0Values(Element ph) {
String mbrTargets = ph.getAttribute("f0");
if (mbrTargets.equals("")) {
return new Float[0];
}
ArrayList<Float> values = new ArrayList<Float>();
try {
// mbrTargets contains one or more pairs of numbers,
// either enclosed by (a,b) or just separated by whitespace.
StringTokenizer st = new StringTokenizer(mbrTargets, " (,)");
while (st.hasMoreTokens()) {
String posString = "";
while (st.hasMoreTokens() && posString.equals(""))
posString = st.nextToken();
String f0String = "";
while (st.hasMoreTokens() && f0String.equals(""))
f0String = st.nextToken();
float pos = Float.parseFloat(posString) * 0.01f;
assert 0 <= pos && pos <= 1 : "invalid position:" + pos + " (pos string was '" + posString
+ "' coming from '" + mbrTargets + "')";
float f0 = Float.parseFloat(f0String);
float logF0 = (float) Math.log(f0);
values.add(pos);
values.add(logF0);
}
} catch (Exception e) {
return new Float[0];
}
return values.toArray(new Float[0]);
}
private float getDuration(Element ph) {
float phoneDuration = 0;
String sDur = ph.getAttribute("d");
if (!sDur.equals("")) {
try {
phoneDuration = Float.parseFloat(sDur);
} catch (NumberFormatException nfe) {
}
}
return phoneDuration;
}
}
/**
* Calculates the slope of a linear approximation of the fundamental frequency, in the log domain. The slope is computed by
* linearly connecting the two log f0 values closest to the middle of the unit segment. This processor should be used by
* target items only -- for unit features during voice building, the actual measured values should be used.
*/
public static class UnitLogF0Delta extends UnitLogF0 {
@Override
public String getName() {
return "unit_logf0delta";
}
public float process(Target target) {
return process(target, true);
}
}
/**
* Returns the value of the given feature for the given segment.
*/
public static class GenericContinuousFeature implements ContinuousFeatureProcessor {
private String name;
private String attributeName;
public GenericContinuousFeature(String featureName, String attributeName) {
this.name = featureName;
this.attributeName = attributeName;
}
public String getName() {
return name;
}
public float process(Target target) {
if (target instanceof DiphoneTarget) {
DiphoneTarget diphone = (DiphoneTarget) target;
// return mean of left and right costs:
return (process(diphone.left) + process(diphone.right)) / 2.0f;
}
Element seg = target.getMaryxmlElement();
if (seg == null) {
return 0;
}
float value = 0;
String valueString;
if (seg.getTagName().equals(MaryXML.PHONE)) {
valueString = seg.getAttribute(attributeName);
} else {
assert seg.getTagName().equals(MaryXML.BOUNDARY) : "segment should be a phone or a boundary, but is a "
+ seg.getTagName();
valueString = seg.getAttribute(attributeName);
}
if (valueString.equals("")) {
return 0;
}
try {
value = Float.parseFloat(valueString);
} catch (NumberFormatException nfe) {
return 0;
}
return value;
}
}
}