Merge pull request #147 from grammatek/master

Merge v1.3.x into master
grammatek · Jan 23, 2024 · e4a8f9d · e4a8f9d
2 parents 28a53b6 + 56d7141
commit e4a8f9d
Show file tree

Hide file tree

Showing 6 changed files with 45 additions and 32 deletions.
diff --git a/app/src/main/java/com/grammatek/simaromur/AppRepository.java b/app/src/main/java/com/grammatek/simaromur/AppRepository.java
@@ -28,6 +28,7 @@
 import com.grammatek.simaromur.db.Voice;
 import com.grammatek.simaromur.db.VoiceDao;
 import com.grammatek.simaromur.device.DownloadVoiceManager;
+import com.grammatek.simaromur.device.SymbolsLvLIs;
 import com.grammatek.simaromur.device.TTSAudioControl;
 import com.grammatek.simaromur.device.TTSEngineController;
 import com.grammatek.simaromur.device.pojo.DeviceVoice;
@@ -901,8 +902,10 @@ public CacheItem executeFrontendAndSaveIntoCache(String text, CacheItem item, co
             // we always need to normalize the text, but it doesn't hurt, if we always do G2P as well
             // for network voices, this is currently all that is needed. But there is an audible
             // problem with trailing "." though, so we remove it
-            final String normalizedText = mFrontend.getNormalizationManager().process(text).replaceAll("\\.+$", "");
+            String normalizedText = mFrontend.getNormalizationManager().process(text).replaceAll("\\.+$", "");
             final String phonemes = mFrontend.transcribe(normalizedText, voice.type, voice.version);
+            // prevent the network voices from pronouncing 'sil'
+            normalizedText = normalizedText.replaceAll(SymbolsLvLIs.TagPause, ",");
             Log.v(LOG_TAG, "onSynthesizeText: original (\"" + text + "\"), normalized (\"" + normalizedText + "\"), phonemes (\"" + phonemes + "\")");
             Utterance updatedUtterance = UtteranceCacheManager.newUtterance(text, normalizedText, List.of(phonemes));
             item = mUtteranceCacheManager.saveUtterance(updatedUtterance);

diff --git a/app/src/main/java/com/grammatek/simaromur/device/SymbolsLvLIs.java b/app/src/main/java/com/grammatek/simaromur/device/SymbolsLvLIs.java
@@ -47,6 +47,9 @@ public enum Type {
     private final static String SymbolsSpecial = SymbolShortPause + " " + SymbolSpokenNoise + " " +
             SymbolSilence;
 
+    // tags
+    public final static String TagPause = "<sil>";
+
     // IPA symbols as HashMap
     private static final HashMap<String, Integer> IPASymbolMap;
     static {

diff --git a/app/src/main/java/com/grammatek/simaromur/frontend/NormalizationManager.java b/app/src/main/java/com/grammatek/simaromur/frontend/NormalizationManager.java
@@ -3,6 +3,8 @@
 import android.content.Context;
 import android.util.Log;
 
+import com.grammatek.simaromur.device.SymbolsLvLIs;
+
 import opennlp.tools.postag.POSModel;
 import opennlp.tools.postag.POSTaggerME;
 
@@ -75,11 +77,11 @@ private List<String> normalize(final List<String> tokenized) {
             // Some very basic phrasing for longer sentences
             // TODO: improve!
             if (tags.length >= 10) {
-                postNormalized = postNormalized.replace(" og ", " <sil> og ");
-                postNormalized = postNormalized.replace(" en ", " <sil> en ");
-                postNormalized = postNormalized.replace(" þegar ", " <sil> þegar ");
-                postNormalized = postNormalized.replace(" sem ", " <sil> sem ");
-                postNormalized = postNormalized.replace(" ef ", " <sil> ef ");
+                postNormalized = postNormalized.replace(" og ", " " + SymbolsLvLIs.TagPause + " og ");
+                postNormalized = postNormalized.replace(" en ", " " + SymbolsLvLIs.TagPause + " en ");
+                postNormalized = postNormalized.replace(" þegar ", " " + SymbolsLvLIs.TagPause + " þegar ");
+                postNormalized = postNormalized.replace(" sem ", " " + SymbolsLvLIs.TagPause + " sem ");
+                postNormalized = postNormalized.replace(" ef ", " " + SymbolsLvLIs.TagPause + " ef ");
             }
             normalized.add(postNormalized);
         }

diff --git a/app/src/main/java/com/grammatek/simaromur/frontend/Pronunciation.java b/app/src/main/java/com/grammatek/simaromur/frontend/Pronunciation.java
@@ -129,7 +129,6 @@ public String convert(String transcribed, String fromAlphabet, String toAlphabet
 
     @NonNull
     private String transcribeString(String text, boolean isFlitev02) {
-        final String silToken = "<sil>";
         String[] tokens = text.split(" ");
         StringBuilder sb = new StringBuilder();
         for (String tok : tokens) {
@@ -139,7 +138,7 @@ private String transcribeString(String text, boolean isFlitev02) {
             else if (mPronDict.containsKey(tok)) {
                 transcr = mPronDict.get(tok).getTranscript().trim();
             }
-            else if (tok.equals(silToken)){
+            else if (tok.equals(SymbolsLvLIs.TagPause)){
                 transcr = SymbolsLvLIs.SymbolShortPause;
             }
             else {
@@ -154,6 +153,10 @@ else if (tok.equals(silToken)){
             if (isFlitev02 && transcr.matches(".+s I n s"))
                 transcr = transcr.replaceAll("s I n s", "s I n n s");
 
+            // bug in Thrax grammar, catch the error here: insert space before C if missing
+            // like in 'Vilhjálmsdóttur' -> 'v I lC au l m s t ou h t Y r'
+            // TODO: remove when Thrax grammar is fixed!
+            transcr = transcr.replaceAll("([a-zA-Z])C", "$1 C");
             sb.append(transcr).append(" ");
         }
         return sb.toString().trim();

diff --git a/app/src/main/java/com/grammatek/simaromur/frontend/TTSNormalizer.java b/app/src/main/java/com/grammatek/simaromur/frontend/TTSNormalizer.java
@@ -4,6 +4,8 @@
 
 import androidx.annotation.NonNull;
 
+import com.grammatek.simaromur.device.SymbolsLvLIs;
+
 import java.util.*;
 import java.util.regex.Matcher;
 import java.util.regex.Pattern;
@@ -297,7 +299,7 @@ else if (numberToken.matches(NumberHelper.FRACTION_PTRN)) {
                 part2 = normalizeThousandDigit(part2, nextTag);
             else
                 part2 = normalizeNumber(part2, nextTag);
-            normalized = part1 + " <sil> " + part2;
+            normalized = part1 + " " + SymbolsLvLIs.TagPause + " " + part2;
         }
         // 01. (what kind of ordinal is this?)
         else if (numberToken.matches("^0\\d\\.$")) {
@@ -473,7 +475,7 @@ private String normalizeDigitOrdinal(String token) {
      * nominative representation, also  '+', '/', ':' are replaced.
      */
     private String normalizeDigits(String token) {
-        token = token.replaceAll(" ", "<sil> ");
+        token = token.replaceAll(" ", SymbolsLvLIs.TagPause + " ");
         for (String digit : NumberHelper.DIGIT_NUMBERS.keySet()) {
             final String replacement = NumberHelper.DIGIT_NUMBERS.get(digit);
             if (replacement != null) {