Merge pull request #44 from idiap/phoneme-cleaners

Add multilingual phoneme cleaner

TTS/tts/utils/text/cleaners.py

@@ -3,6 +3,7 @@
 # TODO: pick the cleaner for languages dynamically
 
 import re
+from typing import Optional
 
 from anyascii import anyascii
 
@@ -17,35 +18,38 @@
 _whitespace_re = re.compile(r"\s+")
 
 
-def expand_abbreviations(text, lang="en"):
+def expand_abbreviations(text: str, lang: str = "en") -> str:
     if lang == "en":
         _abbreviations = abbreviations_en
     elif lang == "fr":
         _abbreviations = abbreviations_fr
+    else:
+        msg = f"Language {lang} not supported in expand_abbreviations"
+        raise ValueError(msg)
     for regex, replacement in _abbreviations:
         text = re.sub(regex, replacement, text)
     return text
 
 
-def lowercase(text):
+def lowercase(text: str) -> str:
     return text.lower()
 
 
-def collapse_whitespace(text):
+def collapse_whitespace(text: str) -> str:
     return re.sub(_whitespace_re, " ", text).strip()
 
 
-def convert_to_ascii(text):
+def convert_to_ascii(text: str) -> str:
     return anyascii(text)
 
 
-def remove_aux_symbols(text):
+def remove_aux_symbols(text: str) -> str:
     text = re.sub(r"[\<\>\(\)\[\]\"]+", "", text)
     return text
 
 
-def replace_symbols(text, lang="en"):
-    """Replace symbols based on the lenguage tag.
+def replace_symbols(text: str, lang: Optional[str] = "en") -> str:
+    """Replace symbols based on the language tag.
 
     Args:
       text:
@@ -77,38 +81,38 @@ def replace_symbols(text, lang="en"):
     return text
 
 
-def basic_cleaners(text):
+def basic_cleaners(text: str) -> str:
     """Basic pipeline that lowercases and collapses whitespace without transliteration."""
     text = lowercase(text)
     text = collapse_whitespace(text)
     return text
 
 
-def transliteration_cleaners(text):
+def transliteration_cleaners(text: str) -> str:
     """Pipeline for non-English text that transliterates to ASCII."""
     # text = convert_to_ascii(text)
     text = lowercase(text)
     text = collapse_whitespace(text)
     return text
 
 
-def basic_german_cleaners(text):
+def basic_german_cleaners(text: str) -> str:
     """Pipeline for German text"""
     text = lowercase(text)
     text = collapse_whitespace(text)
     return text
 
 
 # TODO: elaborate it
-def basic_turkish_cleaners(text):
+def basic_turkish_cleaners(text: str) -> str:
     """Pipeline for Turkish text"""
     text = text.replace("I", "ı")
     text = lowercase(text)
     text = collapse_whitespace(text)
     return text
 
 
-def english_cleaners(text):
+def english_cleaners(text: str) -> str:
     """Pipeline for English text, including number and abbreviation expansion."""
     # text = convert_to_ascii(text)
     text = lowercase(text)
@@ -121,8 +125,12 @@ def english_cleaners(text):
     return text
 
 
-def phoneme_cleaners(text):
-    """Pipeline for phonemes mode, including number and abbreviation expansion."""
+def phoneme_cleaners(text: str) -> str:
+    """Pipeline for phonemes mode, including number and abbreviation expansion.
+
+    NB: This cleaner converts numbers into English words, for other languages
+    use multilingual_phoneme_cleaners().
+    """
     text = en_normalize_numbers(text)
     text = expand_abbreviations(text)
     text = replace_symbols(text)
@@ -131,7 +139,15 @@ def phoneme_cleaners(text):
     return text
 
 
-def french_cleaners(text):
+def multilingual_phoneme_cleaners(text: str) -> str:
+    """Pipeline for phonemes mode, including number and abbreviation expansion."""
+    text = replace_symbols(text, lang=None)
+    text = remove_aux_symbols(text)
+    text = collapse_whitespace(text)
+    return text
+
+
+def french_cleaners(text: str) -> str:
     """Pipeline for French text. There is no need to expand numbers, phonemizer already does that"""
     text = expand_abbreviations(text, lang="fr")
     text = lowercase(text)
@@ -141,7 +157,7 @@ def french_cleaners(text):
     return text
 
 
-def portuguese_cleaners(text):
+def portuguese_cleaners(text: str) -> str:
     """Basic pipeline for Portuguese text. There is no need to expand abbreviation and
     numbers, phonemizer already does that"""
     text = lowercase(text)
@@ -157,7 +173,7 @@ def chinese_mandarin_cleaners(text: str) -> str:
     return text
 
 
-def multilingual_cleaners(text):
+def multilingual_cleaners(text: str) -> str:
     """Pipeline for multilingual text"""
     text = lowercase(text)
     text = replace_symbols(text, lang=None)
@@ -166,7 +182,7 @@ def multilingual_cleaners(text):
     return text
 
 
-def no_cleaners(text):
+def no_cleaners(text: str) -> str:
     # remove newline characters
     text = text.replace("\n", "")
     return text

recipes/thorsten_DE/align_tts/train_aligntts.py

@@ -30,7 +30,7 @@
     run_eval=True,
     test_delay_epochs=-1,
     epochs=1000,
-    text_cleaner="phoneme_cleaners",
+    text_cleaner="multilingual_phoneme_cleaners",
     use_phonemes=False,
     phoneme_language="de",
     phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),

recipes/thorsten_DE/glow_tts/train_glowtts.py

@@ -40,7 +40,7 @@
     run_eval=True,
     test_delay_epochs=-1,
     epochs=1000,
-    text_cleaner="phoneme_cleaners",
+    text_cleaner="multilingual_phoneme_cleaners",
     use_phonemes=True,
     phoneme_language="de",
     phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),

recipes/thorsten_DE/speedy_speech/train_speedy_speech.py

@@ -45,7 +45,7 @@
     test_delay_epochs=-1,
     epochs=1000,
     min_audio_len=11050,  # need to up min_audio_len to avois speedy speech error
-    text_cleaner="phoneme_cleaners",
+    text_cleaner="multilingual_phoneme_cleaners",
     use_phonemes=True,
     phoneme_language="de",
     phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),

recipes/thorsten_DE/tacotron2-DDC/train_tacotron_ddc.py

@@ -49,7 +49,7 @@
     gradual_training=[[0, 6, 64], [10000, 4, 32], [50000, 3, 32], [100000, 2, 32]],
     double_decoder_consistency=True,
     epochs=1000,
-    text_cleaner="phoneme_cleaners",
+    text_cleaner="multilingual_phoneme_cleaners",
     use_phonemes=True,
     phoneme_language="de",
     phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),

recipes/thorsten_DE/vits_tts/train_vits.py

@@ -40,7 +40,7 @@
     run_eval=True,
     test_delay_epochs=-1,
     epochs=1000,
-    text_cleaner="phoneme_cleaners",
+    text_cleaner="multilingual_phoneme_cleaners",
     use_phonemes=True,
     phoneme_language="de",
     phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),

tests/text_tests/test_text_cleaners.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 
-from TTS.tts.utils.text.cleaners import english_cleaners, phoneme_cleaners
+from TTS.tts.utils.text.cleaners import english_cleaners, multilingual_phoneme_cleaners, phoneme_cleaners
 
 
 def test_time() -> None:
@@ -19,3 +19,8 @@ def test_currency() -> None:
 def test_expand_numbers() -> None:
     assert phoneme_cleaners("-1") == "minus one"
     assert phoneme_cleaners("1") == "one"
+
+
+def test_multilingual_phoneme_cleaners() -> None:
+    assert multilingual_phoneme_cleaners("(Hello)") == "Hello"
+    assert multilingual_phoneme_cleaners("1:") == "1,"