package marytts.tools.voiceimport;
import java.io.BufferedReader;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileNotFoundException;
import java.io.FileReader;
import java.io.FileWriter;
import java.io.IOException;
import java.io.InputStreamReader;
import java.util.HashMap;
import java.util.Map;
import java.util.Scanner;
import java.util.StringTokenizer;
import org.apache.commons.io.FileUtils;
/**
* @author marc, marcela
*
*/
public class HMMVoiceCompiler extends VoiceCompiler {
// constants to access filenames in database component properties and organize file list:
/**
* HMM Voice-specific parameters, these are parameters used during models training if using MGC: gamma=0 alpha=0.42 linear
* gain (default) if using LSP: gamma>0 LSP: gamma=1 alpha=0.0 linear gain/log gain Mel-LSP: gamma=1 alpha=0.42 log gain
* MGC-LSP: gamma=3 alpha=0.42 log gain
*/
public static final String alpha = "HMMVoiceConfigure.freqWarp";
public static final String gamma = "HMMVoiceConfigure.gamma";
public static final String logGain = "HMMVoiceConfigure.lnGain";
/** Sampling frequency and frame period have to be specified (sampling freq is included in the general config) */
public static final String samplingRate = "HMMVoiceConfigure.sampFreq";
public static final String framePeriod = "HMMVoiceConfigure.frameShift";
/** The following files depend on the version number and question number defined during configuration and training */
public static final String questionNumber = "HMMVoiceConfigure.questionsNum";
public static final String versionNumber = "HMMVoiceConfigure.version";
/** Tree files and TreeSet object */
public static String treeDurFile;
public static String treeLf0File;
public static String treeMcpFile;
public static String treeStrFile;
/** HMM pdf model files and ModelSet object */
public static String pdfDurFile;
public static String pdfLf0File;
public static String pdfMcpFile;
public static String pdfStrFile;
/** Global variance files */
public static String pdfLf0GvFile;
public static String pdfMcpGvFile;
public static String pdfStrGvFile;
/** Variables for mixed excitation */
public static final String mixFiltersFile = "HMMVoiceConfigure.strFilterFileName";
public static String mixFiltersFileLocation;
public static final String numFilters = "HMMVoiceConfigure.strOrder";
/** Example context feature file (TARGETFEATURES in MARY) */
public static String featuresFileExample = "mary/features_example.pfeats";
public String FeaFile;
/**
* trickyPhones file if any, this file could have been created during makeQuestions and makeLabels if it was created, because
* there are tricky phones in the allophones set, then it should be in voiceDIR/mary/trickyPhones.txt
*/
public static final String trickyPhonesFile = "HMMVoiceMakeData.trickyPhonesFile";
public static final String hmmFeaturesMapFile = "HMMVoiceMakeData.featureListMapFile";
/** Mapping in case of using alias names for extra features during training */
Map<String, String> actualFeatureNames = new HashMap<String, String>();
/**
*
*/
public HMMVoiceCompiler() {
}
/**
* @throws IOException
* IOException
* @throws FileNotFoundException
* FileNotFoundException
*/
@Override
protected void mapFeatures() throws IOException, FileNotFoundException {
String rootDir = db.getProp(DatabaseLayout.ROOTDIR);
// First find a features file example
getFeatureFileExample();
/* Substitute question number and version number */
String vnum = db.getProperty(versionNumber);
String qnum = db.getProperty(questionNumber);
/** Tree files and TreeSet object */
treeDurFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/tree-dur.inf";
treeLf0File = "hts/voices/qst" + qnum + "/ver" + vnum + "/tree-lf0.inf";
treeMcpFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/tree-mgc.inf";
treeStrFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/tree-str.inf";
/** HMM pdf model files and ModelSet object */
pdfDurFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/dur.pdf";
pdfLf0File = "hts/voices/qst" + qnum + "/ver" + vnum + "/lf0.pdf";
pdfMcpFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/mgc.pdf";
pdfStrFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/str.pdf";
/** Global variance files */
pdfLf0GvFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/gv-lf0.pdf";
pdfMcpGvFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/gv-mgc.pdf";
pdfStrGvFile = "hts/voices/qst" + qnum + "/ver" + vnum + "/gv-str.pdf";
/** Filter file for mixed excitation */
mixFiltersFileLocation = "hts/data/" + db.getProperty(mixFiltersFile);
// Set files for resources
// Now I know the names of the resources so I can set the files for resources on the maven compiler
String[] filenamesResources = new String[] { rootDir + treeDurFile, rootDir + treeLf0File, rootDir + treeMcpFile,
rootDir + treeStrFile, rootDir + pdfDurFile, rootDir + pdfLf0File, rootDir + pdfMcpFile, rootDir + pdfStrFile,
rootDir + pdfLf0GvFile, rootDir + pdfMcpGvFile, rootDir + pdfStrGvFile, rootDir + mixFiltersFileLocation,
rootDir + featuresFileExample, rootDir + db.getProperty(trickyPhonesFile) };
File[] filesForResources = new File[filenamesResources.length];
for (int i = 0; i < filenamesResources.length; i++) {
filesForResources[i] = new File(filenamesResources[i]);
}
compiler.setFilesForResources(filesForResources);
// Before setting the tree files, we need to check if they contain aliases for the extra features used for training
// if so there must be a file mary/hmmFeaturesMap.txt which has to be used to convert back the feature names
// Check if features map was used
String feasMapFileName = rootDir + db.getProperty(hmmFeaturesMapFile);
System.out.println("Checking if aliases for extra features used for training were used: checking if file exist -->"
+ feasMapFileName);
File featuresMap = new File(feasMapFileName);
if (featuresMap.exists()) {
// convert back the features in all tree files: treeDurFile, treeLf0File, treeMcpFile, treeStrFile
System.out.println("convert back the features in all tree files: treeDurFile, treeLf0File, treeMcpFile, treeStrFile");
loadFeaturesMap(feasMapFileName);
replaceBackFeatureNames(rootDir + treeDurFile);
replaceBackFeatureNames(rootDir + treeLf0File);
replaceBackFeatureNames(rootDir + treeMcpFile);
replaceBackFeatureNames(rootDir + treeStrFile);
}
}
@Override
protected boolean isUnitSelectionVoice() {
return false;
}
@Override
protected String getVoiceName(DatabaseLayout db) {
return db.getVoiceName() + "-hsmm";
}
/*
* (non-Javadoc)
*
* @see marytts.tools.voiceimport.VoiceImportComponent#getName()
*/
@Override
public String getName() {
return "HMMVoiceCompiler";
}
@Override
protected Map<String, String> getExtraVariableSubstitutionMap() {
Map<String, String> m = new HashMap<String, String>();
m.put("FRAMEPERIOD", String.valueOf(db.getProperty(framePeriod)));
m.put("ALPHA", String.valueOf(db.getProperty(alpha)));
m.put("GAMMA", String.valueOf(db.getProperty(gamma)));
m.put("LOGGAIN", String.valueOf(db.getProperty(logGain)));
// The filters file name includes the "filters/" directory, we need here just the file name
m.put("MIXEXCFILTERFILE", db.getProperty(mixFiltersFile).substring(8));
m.put("NUMMIXEXCFILTERS", String.valueOf(db.getProperty(numFilters)));
return m;
}
@Override
protected File[] getFilesForResources() {
// at this point, initialisation, I still do not know where
// exactly the files will be
/*
* String[] filenamesResources = new String[] { treeDurFile, treeLf0File, treeMcpFile, treeStrFile, pdfDurFile,
* pdfLf0File, pdfMcpFile, pdfStrFile, pdfLf0GvFile, pdfMcpGvFile, pdfStrGvFile, mixFiltersFileLocation,
* featuresFileExample, db.getProperty(trickyPhonesFile) }; File[] filesForResources = new
* File[filenamesResources.length]; for (int i=0; i<filenamesResources.length; i++) { filesForResources[i] = new
* File(filenamesResources[i]); } return filesForResources;
*/
return null;
}
@Override
protected File[] getFilesForFilesystem() {
return null;
}
protected void getFeatureFileExample() throws IOException {
String fileExample = null;
String rootDir = db.getProp(DatabaseLayout.ROOTDIR);
logger.info("Getting a context feature file example in phonefeatures/");
/*
* copy one example of MARY context features file, it can be one of the files used for testing in phonefeatures/*.pfeats
*/
File dirPhonefeatures = new File(rootDir + "phonefeatures/");
if (dirPhonefeatures.exists() && dirPhonefeatures.list().length > 0) {
String[] feaFiles = dirPhonefeatures.list();
fileExample = feaFiles[0];
File in = new File(rootDir + "phonefeatures/" + fileExample);
if (in.isDirectory()) {
logger.info("HMMVoiceConfigure.adaptScripts = " + db.getProperty("HMMVoiceConfigure.adaptScripts"));
/* If adaptive training then look for an example in the first directory */
if (in.exists() && in.list().length > 0) {
FileUtils.copyFile(new File(rootDir, "phonefeatures/" + fileExample + "/" + in.list()[0]), new File(rootDir,
featuresFileExample));
}
} else if (in.exists()) {
FileUtils.copyFile(in, new File(rootDir, featuresFileExample));
} else {
System.out.println("Problem copying one example of context features, the directory phonefeatures/ is empty(?)");
throw new IOException();
}
} else {
System.out.println("Problem copying one example of context features, the directory does not exist.");
throw new IOException();
}
}
/**
* Replace the aliases for features used during training
*
* @param treeFileName
* a HTS tree file
* @throws IOException
* IOException
*/
private void replaceBackFeatureNames(String treeFileName) throws IOException {
BufferedReader s = null;
FileWriter outputStream;
// ---outputStream.write(hmm_tts.getRealisedDurations());
// ---outputStream.close();
String line;
// read the file until the symbol the delimits an state is found
try {
// output file to copy the result
outputStream = new FileWriter(treeFileName + ".tmp");
// read lines of tree-*.inf fileName
s = new BufferedReader(new InputStreamReader(new FileInputStream(treeFileName)));
logger.info("load: reading " + treeFileName);
// skip questions section, but copy the lines on the temporary output
while ((line = s.readLine()) != null) {
outputStream.write(line + "\n");
if (line.indexOf("QS") < 0)
break; /* a new state is indicated by {*}[2], {*}[3], ... */
}
StringTokenizer sline;
String buf1, buf2, buf3, buf4;
while ((line = s.readLine()) != null) {
// System.out.println("line: " + line);
if (line.indexOf("{") >= 0 || line.indexOf("}") >= 0 || line.length() == 0) { /*
* this is the indicator of a new
* state-tree
*/
outputStream.write(line + "\n");
} else {
sline = new StringTokenizer(line);
buf1 = sline.nextToken();
buf2 = sline.nextToken();
String[] fea = buf2.split("=");
buf3 = sline.nextToken();
buf4 = sline.nextToken();
// System.out.format("newLine: %s %s=%s\t\t%s\t%s\n", buf1, replaceBack(fea[0]), fea[1], buf3, buf4);
outputStream
.write(" " + buf1 + " " + replaceBack(fea[0]) + "=" + fea[1] + "\t\t" + buf3 + "\t" + buf4 + "\n");
}
}
outputStream.close();
System.out.println("Features alises replaced in file: " + treeFileName + ".tmp");
// now replace the file
FileUtils.copyFile(new File(treeFileName + ".tmp"), new File(treeFileName));
System.out.println("Copied file: " + treeFileName + ".tmp" + " to: " + treeFileName + "\n");
} catch (IOException e) {
logger.debug("FileNotFoundException: " + e.getMessage());
throw new IOException("LoadTreeSet: ", e);
}
}
/**
* Load mapping of features from file
*
* @param fileName
* fileName
* @throws FileNotFoundException
* FileNotFoundException
*/
private Map<String, String> loadFeaturesMap(String fileName) throws FileNotFoundException {
Scanner aliasList = null;
try {
aliasList = new Scanner(new BufferedReader(new FileReader(fileName)));
String line;
logger.info("loading features map from file: " + fileName);
while (aliasList.hasNext()) {
line = aliasList.nextLine();
String[] fea = line.split(" ");
actualFeatureNames.put(fea[1], fea[0]);
logger.info(" " + fea[0] + " --> " + fea[1]);
}
if (aliasList != null) {
aliasList.close();
}
} catch (FileNotFoundException e) {
logger.debug("loadTrickyPhones: " + e.getMessage());
throw e;
}
return actualFeatureNames;
}
/**
* Replace label with information in the global map list actualFeatureNames
*
* @param lab
* replaced label
* @return s
*/
public String replaceBack(String lab) {
String s = lab;
if (actualFeatureNames.containsKey(lab)) {
s = actualFeatureNames.get(lab);
}
return s;
}
}