Upgrade to gruut 2.0 (#882)

TTS/tts/utils/text/__init__.py

@@ -5,6 +5,7 @@
 from typing import Dict, List
 
 import gruut
+from gruut_ipa import IPA
 
 from TTS.tts.utils.text import cleaners
 from TTS.tts.utils.text.chinese_mandarin.phonemizer import chinese_text_to_phonemes
@@ -32,7 +33,7 @@
 GRUUT_TRANS_TABLE = str.maketrans("g", "ɡ")
 
 
-def text2phone(text, language, use_espeak_phonemes=False):
+def text2phone(text, language, use_espeak_phonemes=False, keep_stress=False):
     """Convert graphemes to phonemes.
     Parameters:
             text (str): text to phonemize
@@ -51,36 +52,44 @@ def text2phone(text, language, use_espeak_phonemes=False):
         ph = japanese_text_to_phonemes(text)
         return ph
 
-    if gruut.is_language_supported(language):
-        # Use gruut for phonemization
-        phonemizer_args = {
-            "remove_stress": True,
-            "ipa_minor_breaks": False,  # don't replace commas/semi-colons with IPA |
-            "ipa_major_breaks": False,  # don't replace periods with IPA ‖
-        }
-
-        if use_espeak_phonemes:
-            # Use a lexicon/g2p model train on eSpeak IPA instead of gruut IPA.
-            # This is intended for backwards compatibility with TTS<=v0.0.13
-            # pre-trained models.
-            phonemizer_args["model_prefix"] = "espeak"
-
-        ph_list = gruut.text_to_phonemes(
-            text,
-            lang=language,
-            return_format="word_phonemes",
-            phonemizer_args=phonemizer_args,
-        )
-
-        # Join and re-split to break apart dipthongs, suprasegmentals, etc.
-        ph_words = ["|".join(word_phonemes) for word_phonemes in ph_list]
-        ph = "| ".join(ph_words)
-
-        # Fix a few phonemes
-        ph = ph.translate(GRUUT_TRANS_TABLE)
-        return ph
-
-    raise ValueError(f" [!] Language {language} is not supported for phonemization.")
+    if not gruut.is_language_supported(language):
+        raise ValueError(f" [!] Language {language} is not supported for phonemization.")
+
+    # Use gruut for phonemization
+    ph_list = []
+    for sentence in gruut.sentences(text, lang=language, espeak=use_espeak_phonemes):
+        for word in sentence:
+            if word.is_break:
+                # Use actual character for break phoneme (e.g., comma)
+                if ph_list:
+                    # Join with previous word
+                    ph_list[-1].append(word.text)
+                else:
+                    # First word is punctuation
+                    ph_list.append([word.text])
+            elif word.phonemes:
+                # Add phonemes for word
+                word_phonemes = []
+
+                for word_phoneme in word.phonemes:
+                    if not keep_stress:
+                        # Remove primary/secondary stress
+                        word_phoneme = IPA.without_stress(word_phoneme)
+
+                    word_phoneme = word_phoneme.translate(GRUUT_TRANS_TABLE)
+
+                    if word_phoneme:
+                        # Flatten phonemes
+                        word_phonemes.extend(word_phoneme)
+
+                if word_phonemes:
+                    ph_list.append(word_phonemes)
+
+    # Join and re-split to break apart dipthongs, suprasegmentals, etc.
+    ph_words = ["|".join(word_phonemes) for word_phonemes in ph_list]
+    ph = "| ".join(ph_words)
+
+    return ph
 
 
 def intersperse(sequence, token):

requirements.txt

@@ -23,6 +23,6 @@ coqpit
 mecab-python3==1.0.3
 unidic-lite==1.0.8
 # gruut+supported langs
-gruut[cs,de,es,fr,it,nl,pt,ru,sv]~=1.2.0
+gruut[cs,de,es,fr,it,nl,pt,ru,sv]~=2.0.0
 fsspec>=2021.04.0
-pyworld
+pyworld

tests/aux_tests/test_text_processing.py

@@ -9,12 +9,12 @@
 
 EXAMPLE_TEXT = "Recent research at Harvard has shown meditating for as little as 8 weeks can actually increase, the grey matter in the parts of the brain responsible for emotional regulation and learning!"
 
-EXPECTED_PHONEMES = "ɹ|iː|s|ə|n|t| ɹ|ᵻ|s|ɜː|tʃ| æ|t| h|ɑːɹ|v|ɚ|d| h|æ|z| ʃ|oʊ|n| m|ɛ|d|ᵻ|t|eɪ|ɾ|ɪ|ŋ| f|ɔːɹ| æ|z| l|ɪ|ɾ|əl| æ|z| eɪ|t| w|iː|k|s| k|æ|n| æ|k|tʃ|uː|əl|i| ɪ|ŋ|k|ɹ|iː|s| ,| ð|ə| ɡ|ɹ|eɪ| m|æ|ɾ|ɚ| ɪ|n| ð|ə| p|ɑːɹ|t|s| ʌ|v| ð|ə| b|ɹ|eɪ|n| ɹ|ᵻ|s|p|ɑː|n|s|ᵻ|b|əl| f|ɔːɹ| ɪ|m|oʊ|ʃ|ə|n|əl| ɹ|ɛ|ɡ|j|ʊ|l|eɪ|ʃ|ə|n| æ|n|d| l|ɜː|n|ɪ|ŋ| !"
+EXPECTED_PHONEMES = "ɹ|i|ː|s|ə|n|t| ɹ|ᵻ|s|ɜ|ː|t|ʃ| æ|ɾ| h|ɑ|ː|ɹ|v|ɚ|d| h|ɐ|z| ʃ|o|ʊ|n| m|ɛ|d|ᵻ|t|e|ɪ|ɾ|ɪ|ŋ| f|ɔ|ː|ɹ| æ|z| l|ɪ|ɾ|ə|l| æ|z| e|ɪ|t| w|i|ː|k|s| k|æ|ŋ| æ|k|t|ʃ|u|ː|ə|l|i| ɪ|ŋ|k|ɹ|i|ː|s|,| ð|ə| ɡ|ɹ|e|ɪ| m|æ|ɾ|ɚ| ɪ|n| ð|ə| p|ɑ|ː|ɹ|t|s| ʌ|v| ð|ə| b|ɹ|e|ɪ|n| ɹ|ᵻ|s|p|ɑ|ː|n|s|ᵻ|b|ə|l| f|ɔ|ː|ɹ| ɪ|m|o|ʊ|ʃ|ə|n|ə|l| ɹ|ɛ|ɡ|j|ʊ|l|e|ɪ|ʃ|ə|n| æ|n|d| l|ɜ|ː|n|ɪ|ŋ|!"
 
 # -----------------------------------------------------------------------------
 
 
-class TextProcessingTextCase(unittest.TestCase):
+class TextProcessingTestCase(unittest.TestCase):
     """Tests for text to phoneme conversion"""
 
     def test_phoneme_to_sequence(self):
@@ -40,7 +40,7 @@ def _test_phoneme_to_sequence(self, add_blank):
         sequence = phoneme_to_sequence(text, text_cleaner, LANG, add_blank=add_blank, use_espeak_phonemes=True)
         text_hat = sequence_to_phoneme(sequence)
         text_hat_with_params = sequence_to_phoneme(sequence)
-        gt = "biː ɐ vɔɪs , nɑːt ɐn ! ɛkoʊ ?"
+        gt = "biː ɐ vɔɪs, nɑːt ɐn! ɛkoʊ?"
         print(text_hat)
         print(len(sequence))
         self.assertEqual(text_hat, text_hat_with_params)
@@ -51,7 +51,7 @@ def _test_phoneme_to_sequence(self, add_blank):
         sequence = phoneme_to_sequence(text, text_cleaner, LANG, add_blank=add_blank, use_espeak_phonemes=True)
         text_hat = sequence_to_phoneme(sequence)
         text_hat_with_params = sequence_to_phoneme(sequence)
-        gt = "biː ɐ vɔɪs , nɑːt ɐn ! ɛkoʊ"
+        gt = "biː ɐ vɔɪs, nɑːt ɐn! ɛkoʊ"
         print(text_hat)
         print(len(sequence))
         self.assertEqual(text_hat, text_hat_with_params)
@@ -62,7 +62,7 @@ def _test_phoneme_to_sequence(self, add_blank):
         sequence = phoneme_to_sequence(text, text_cleaner, LANG, add_blank=add_blank, use_espeak_phonemes=True)
         text_hat = sequence_to_phoneme(sequence)
         text_hat_with_params = sequence_to_phoneme(sequence)
-        gt = "biː ɐ vɔɪs , nɑːt ɐn ɛkoʊ !"
+        gt = "biː ɐ vɔɪs, nɑːt ɐn ɛkoʊ!"
         print(text_hat)
         print(len(sequence))
         self.assertEqual(text_hat, text_hat_with_params)
@@ -73,7 +73,7 @@ def _test_phoneme_to_sequence(self, add_blank):
         sequence = phoneme_to_sequence(text, text_cleaner, LANG, add_blank=add_blank, use_espeak_phonemes=True)
         text_hat = sequence_to_phoneme(sequence)
         text_hat_with_params = sequence_to_phoneme(sequence)
-        gt = "biː ɐ vɔɪs , nɑːt ɐn ! ɛkoʊ ."
+        gt = "biː ɐ vɔɪs, nɑːt ɐn! ɛkoʊ."
         print(text_hat)
         print(len(sequence))
         self.assertEqual(text_hat, text_hat_with_params)
@@ -86,7 +86,7 @@ def _test_phoneme_to_sequence(self, add_blank):
         )
         text_hat = sequence_to_phoneme(sequence)
         text_hat_with_params = sequence_to_phoneme(sequence)
-        gt = "^biː ɐ vɔɪs , nɑːt ɐn ! ɛkoʊ .~"
+        gt = "^biː ɐ vɔɪs, nɑːt ɐn! ɛkoʊ.~"
         print(text_hat)
         print(len(sequence))
         self.assertEqual(text_hat, text_hat_with_params)