ASLP-lab/DiffRhythm

1# Copyright (c) 2024 Amphion.2#3# This source code is licensed under the MIT license found in the4# LICENSE file in the root directory of this source tree.5 6from phonemizer.backend import EspeakBackend7from phonemizer.separator import Separator8from phonemizer.utils import list2str, str2list9from typing import List, Union10import os11import json12import sys13 14# separator=Separator(phone=' ', word=' _ ', syllable='|'),15separator = Separator(word=" _ ", syllable="|", phone=" ")16 17phonemizer_zh = EspeakBackend(18    "cmn", preserve_punctuation=False, with_stress=False, language_switch="remove-flags"19)20# phonemizer_zh.separator = separator21 22phonemizer_en = EspeakBackend(23    "en-us",24    preserve_punctuation=False,25    with_stress=False,26    language_switch="remove-flags",27)28# phonemizer_en.separator = separator29 30phonemizer_ja = EspeakBackend(31    "ja", preserve_punctuation=False, with_stress=False, language_switch="remove-flags"32)33# phonemizer_ja.separator = separator34 35phonemizer_ko = EspeakBackend(36    "ko", preserve_punctuation=False, with_stress=False, language_switch="remove-flags"37)38# phonemizer_ko.separator = separator39 40phonemizer_fr = EspeakBackend(41    "fr-fr",42    preserve_punctuation=False,43    with_stress=False,44    language_switch="remove-flags",45)46# phonemizer_fr.separator = separator47 48phonemizer_de = EspeakBackend(49    "de", preserve_punctuation=False, with_stress=False, language_switch="remove-flags"50)51# phonemizer_de.separator = separator52 53 54lang2backend = {55    "zh": phonemizer_zh,56    "ja": phonemizer_ja,57    "en": phonemizer_en,58    "fr": phonemizer_fr,59    "ko": phonemizer_ko,60    "de": phonemizer_de,61}62 63with open("./diffrhythm/g2p/utils/mls_en.json", "r") as f:64    json_data = f.read()65token = json.loads(json_data)66 67 68def phonemizer_g2p(text, language):69    langbackend = lang2backend[language]70    phonemes = _phonemize(71        langbackend,72        text,73        separator,74        strip=True,75        njobs=1,76        prepend_text=False,77        preserve_empty_lines=False,78    )79    token_id = []80    if isinstance(phonemes, list):81        for phone in phonemes:82            phonemes_split = phone.split(" ")83            token_id.append([token[p] for p in phonemes_split if p in token])84    else:85        phonemes_split = phonemes.split(" ")86        token_id = [token[p] for p in phonemes_split if p in token]87    return phonemes, token_id88 89 90def _phonemize(  # pylint: disable=too-many-arguments91    backend,92    text: Union[str, List[str]],93    separator: Separator,94    strip: bool,95    njobs: int,96    prepend_text: bool,97    preserve_empty_lines: bool,98):99    """Auxiliary function to phonemize()100 101    Does the phonemization and returns the phonemized text. Raises a102    RuntimeError on error.103 104    """105    # remember the text type for output (either list or string)106    text_type = type(text)107 108    # force the text as a list109    text = [line.strip(os.linesep) for line in str2list(text)]110 111    # if preserving empty lines, note the index of each empty line112    if preserve_empty_lines:113        empty_lines = [n for n, line in enumerate(text) if not line.strip()]114 115    # ignore empty lines116    text = [line for line in text if line.strip()]117 118    if text:119        # phonemize the text120        phonemized = backend.phonemize(121            text, separator=separator, strip=strip, njobs=njobs122        )123    else:124        phonemized = []125 126    # if preserving empty lines, reinsert them into text and phonemized lists127    if preserve_empty_lines:128        for i in empty_lines:  # noqa129            if prepend_text:130                text.insert(i, "")131            phonemized.insert(i, "")132 133    # at that point, the phonemized text is a list of str. Format it as134    # expected by the parameters135    if prepend_text:136        return list(zip(text, phonemized))137    if text_type == str:138        return list2str(phonemized)139    return phonemized140