def test_sentence_phonemes(self):
"""Test sentence phonemes return format"""
expected_phonemes = [
["ˈaɪ", "w", "ˈaʊ", "n", "d", "ð", "ə", "w", "ˈu", "n", "d", "‖"],
[
"ˈaɪ",
"ɹ",
"ɪ",
"f",
"j",
"ˈu",
"z",
"ð",
"ə",
"ɹ",
"ˈɛ",
"f",
"j",
"ˌu",
"z",
"‖",
],
]
actual_phonemes = text_to_phonemes(
"I wound the wound. I refuse the refuse.", return_format="sentence_phonemes"
)
self.assertEqual(expected_phonemes, actual_phonemes)
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 test_word_phonemes(self):
"""Test word phonemes return format"""
expected_phonemes = [
["ˈaɪ"],
["w", "ˈaʊ", "n", "d"],
["ð", "ə"],
["w", "ˈu", "n", "d"],
["‖"],
["ˈaɪ"],
["ɹ", "ɪ", "f", "j", "ˈu", "z"],
["ð", "ə"],
["ɹ", "ˈɛ", "f", "j", "ˌu", "z"],
["‖"],
]
actual_phonemes = text_to_phonemes(
"I wound the wound. I refuse the refuse.", return_format="word_phonemes"
)
self.assertEqual(expected_phonemes, actual_phonemes)
def test_sentences(self):
"""Test sentences return format"""
expected_phonemes = [
[["ˈaɪ"], ["w", "ˈaʊ", "n", "d"], ["ð", "ə"], ["w", "ˈu", "n", "d"], ["‖"]],
[
["ˈaɪ"],
["ɹ", "ɪ", "f", "j", "ˈu", "z"],
["ð", "ə"],
["ɹ", "ˈɛ", "f", "j", "ˌu", "z"],
["‖"],
],
]
actual_sentences = text_to_phonemes(
"I wound the wound. I refuse the refuse.", return_format="sentences"
)
# Returns two Sentence objects
self.assertEqual(len(actual_sentences), 2)
self.assertIsInstance(actual_sentences[0], Sentence)
self.assertIsInstance(actual_sentences[1], Sentence)
self.assertEqual(expected_phonemes[0], actual_sentences[0].phonemes)
self.assertEqual(expected_phonemes[1], actual_sentences[1].phonemes)
from gruut import text_to_phonemes
text = 'He wound it around the wound, saying "I read it was $10 to read."'
for sent_idx, word, word_phonemes in text_to_phonemes(text, lang="en-us"):
print(word, *word_phonemes)
def test_last_token(self):
"""Ensure liason does not leave last token"""
phonemes = text_to_phonemes("Est-ce-que", lang="fr")
self.assertGreater(len(phonemes), 0)