def text2phone(text, language, use_espeak_phonemes=False, keep_stress=False):
"""Convert graphemes to phonemes.
Parameters:
text (str): text to phonemize
language (str): language of the text
Returns:
ph (str): phonemes as a string seperated by "|"
ph = "ɪ|g|ˈ|z|æ|m|p|ə|l"
"""
# TO REVIEW : How to have a good implementation for this?
if language == "zh-CN":
ph = chinese_text_to_phonemes(text)
return ph
if language == "ja-jp":
ph = japanese_text_to_phonemes(text)
return ph
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 text2phone(text, language, use_espeak_phonemes=False):
"""Convert graphemes to phonemes.
Parameters:
text (str): text to phonemize
language (str): language of the text
Returns:
ph (str): phonemes as a string seperated by "|"
ph = "ɪ|g|ˈ|z|æ|m|p|ə|l"
"""
# TO REVIEW : How to have a good implementation for this?
if language == "zh-CN":
from TTS.tts.utils.text.chinese_mandarin.phonemizer import (
chinese_text_to_phonemes, )
ph = chinese_text_to_phonemes(text)
return ph
if language == "ja-jp":
from TTS.tts.utils.text.japanese.phonemizer import japanese_text_to_phonemes
ph = japanese_text_to_phonemes(text)
return ph
if gruut.is_language_supported(language):
# 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:
ph_list[-1].append(word.text)
else:
ph_list.append([word.text])
elif word.phonemes:
# Remove stress from word phonemes
word_phonemes = []
for word_phoneme in word.phonemes:
word_phonemes.append(
IPA.without_stress(word_phoneme).translate(
GRUUT_TRANS_TABLE))
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
raise ValueError(
f" [!] Language {language} is not supported for phonemization.")
def text2phone(text, language, use_espeak_phonemes=False):
"""Convert graphemes to phonemes.
Parameters:
text (str): text to phonemize
language (str): language of the text
Returns:
ph (str): phonemes as a string seperated by "|"
ph = "ɪ|g|ˈ|z|æ|m|p|ə|l"
"""
# TO REVIEW : How to have a good implementation for this?
if language == "zh-CN":
ph = chinese_text_to_phonemes(text)
return ph
if language == "ja-jp":
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.")
def text2phone(text, language):
"""Convert graphemes to phonemes.
Parameters:
text (str): text to phonemize
language (str): language of the text
Returns:
ph (str): phonemes as a string seperated by "|"
ph = "ɪ|g|ˈ|z|æ|m|p|ə|l"
"""
# TO REVIEW : How to have a good implementation for this?
if language == "zh-CN":
ph = chinese_text_to_phonemes(text)
return ph
if language == "ja-jp":
ph = japanese_text_to_phonemes(text)
return ph
raise ValueError(f" [!] Language {language} is not supported for phonemization.")
def test_japanese_text_to_phonemes(self):
for line in _TEST_CASES.strip().split("\n"):
text, phone = line.split("/")
self.assertEqual(japanese_text_to_phonemes(text), phone)