|
| 1 | +# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved. |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at |
| 6 | +# |
| 7 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +# See the License for the specific language governing permissions and |
| 13 | +# limitations under the License. |
| 14 | +import re |
| 15 | +from typing import Dict |
| 16 | +from typing import List |
| 17 | + |
| 18 | +import paddle |
| 19 | + |
| 20 | +from paddlespeech.t2s.frontend import English |
| 21 | +from paddlespeech.t2s.frontend.zh_frontend import Frontend |
| 22 | + |
| 23 | + |
| 24 | +class MixFrontend(): |
| 25 | + def __init__(self, |
| 26 | + g2p_model="pypinyin", |
| 27 | + phone_vocab_path=None, |
| 28 | + tone_vocab_path=None): |
| 29 | + |
| 30 | + self.zh_frontend = Frontend( |
| 31 | + phone_vocab_path=phone_vocab_path, tone_vocab_path=tone_vocab_path) |
| 32 | + self.en_frontend = English(phone_vocab_path=phone_vocab_path) |
| 33 | + self.SENTENCE_SPLITOR = re.compile(r'([：、，；。？！,;?!][”’]?)') |
| 34 | + self.sp_id = self.zh_frontend.vocab_phones["sp"] |
| 35 | + self.sp_id_tensor = paddle.to_tensor([self.sp_id]) |
| 36 | + |
| 37 | + def is_chinese(self, char): |
| 38 | + if char >= '\u4e00' and char <= '\u9fa5': |
| 39 | + return True |
| 40 | + else: |
| 41 | + return False |
| 42 | + |
| 43 | + def is_alphabet(self, char): |
| 44 | + if (char >= '\u0041' and char <= '\u005a') or (char >= '\u0061' and |
| 45 | + char <= '\u007a'): |
| 46 | + return True |
| 47 | + else: |
| 48 | + return False |
| 49 | + |
| 50 | + def is_number(self, char): |
| 51 | + if char >= '\u0030' and char <= '\u0039': |
| 52 | + return True |
| 53 | + else: |
| 54 | + return False |
| 55 | + |
| 56 | + def is_other(self, char): |
| 57 | + if not (self.is_chinese(char) or self.is_number(char) or |
| 58 | + self.is_alphabet(char)): |
| 59 | + return True |
| 60 | + else: |
| 61 | + return False |
| 62 | + |
| 63 | + def _split(self, text: str) -> List[str]: |
| 64 | + text = re.sub(r'[《》【】<=>{}()（）#&@“”^_|…\\]', '', text) |
| 65 | + text = self.SENTENCE_SPLITOR.sub(r'\1\n', text) |
| 66 | + text = text.strip() |
| 67 | + sentences = [sentence.strip() for sentence in re.split(r'\n+', text)] |
| 68 | + return sentences |
| 69 | + |
| 70 | + def _distinguish(self, text: str) -> List[str]: |
| 71 | + # sentence --> [ch_part, en_part, ch_part, ...] |
| 72 | + |
| 73 | + segments = [] |
| 74 | + types = [] |
| 75 | + |
| 76 | + flag = 0 |
| 77 | + temp_seg = "" |
| 78 | + temp_lang = "" |
| 79 | + |
| 80 | + # Determine the type of each character. type: blank, chinese, alphabet, number, unk. |
| 81 | + for ch in text: |
| 82 | + if self.is_chinese(ch): |
| 83 | + types.append("zh") |
| 84 | + elif self.is_alphabet(ch): |
| 85 | + types.append("en") |
| 86 | + elif ch == " ": |
| 87 | + types.append("blank") |
| 88 | + elif self.is_number(ch): |
| 89 | + types.append("num") |
| 90 | + else: |
| 91 | + types.append("unk") |
| 92 | + |
| 93 | + assert len(types) == len(text) |
| 94 | + |
| 95 | + for i in range(len(types)): |
| 96 | + |
| 97 | + # find the first char of the seg |
| 98 | + if flag == 0: |
| 99 | + if types[i] != "unk" and types[i] != "blank": |
| 100 | + temp_seg += text[i] |
| 101 | + temp_lang = types[i] |
| 102 | + flag = 1 |
| 103 | + |
| 104 | + else: |
| 105 | + if types[i] == temp_lang or types[i] == "num": |
| 106 | + temp_seg += text[i] |
| 107 | + |
| 108 | + elif temp_lang == "num" and types[i] != "unk": |
| 109 | + temp_seg += text[i] |
| 110 | + if types[i] == "zh" or types[i] == "en": |
| 111 | + temp_lang = types[i] |
| 112 | + |
| 113 | + elif temp_lang == "en" and types[i] == "blank": |
| 114 | + temp_seg += text[i] |
| 115 | + |
| 116 | + elif types[i] == "unk": |
| 117 | + pass |
| 118 | + |
| 119 | + else: |
| 120 | + segments.append((temp_seg, temp_lang)) |
| 121 | + |
| 122 | + if types[i] != "unk" and types[i] != "blank": |
| 123 | + temp_seg = text[i] |
| 124 | + temp_lang = types[i] |
| 125 | + flag = 1 |
| 126 | + else: |
| 127 | + flag = 0 |
| 128 | + temp_seg = "" |
| 129 | + temp_lang = "" |
| 130 | + |
| 131 | + segments.append((temp_seg, temp_lang)) |
| 132 | + |
| 133 | + return segments |
| 134 | + |
| 135 | + def get_input_ids(self, |
| 136 | + sentence: str, |
| 137 | + merge_sentences: bool=True, |
| 138 | + get_tone_ids: bool=False, |
| 139 | + add_sp: bool=True) -> Dict[str, List[paddle.Tensor]]: |
| 140 | + |
| 141 | + sentences = self._split(sentence) |
| 142 | + phones_list = [] |
| 143 | + result = {} |
| 144 | + |
| 145 | + for text in sentences: |
| 146 | + phones_seg = [] |
| 147 | + segments = self._distinguish(text) |
| 148 | + for seg in segments: |
| 149 | + content = seg[0] |
| 150 | + lang = seg[1] |
| 151 | + if lang == "zh": |
| 152 | + input_ids = self.zh_frontend.get_input_ids( |
| 153 | + content, |
| 154 | + merge_sentences=True, |
| 155 | + get_tone_ids=get_tone_ids) |
| 156 | + |
| 157 | + elif lang == "en": |
| 158 | + input_ids = self.en_frontend.get_input_ids( |
| 159 | + content, merge_sentences=True) |
| 160 | + |
| 161 | + phones_seg.append(input_ids["phone_ids"][0]) |
| 162 | + if add_sp: |
| 163 | + phones_seg.append(self.sp_id_tensor) |
| 164 | + |
| 165 | + phones = paddle.concat(phones_seg) |
| 166 | + phones_list.append(phones) |
| 167 | + |
| 168 | + if merge_sentences: |
| 169 | + merge_list = paddle.concat(phones_list) |
| 170 | + # rm the last 'sp' to avoid the noise at the end |
| 171 | + # cause in the training data, no 'sp' in the end |
| 172 | + if merge_list[-1] == self.sp_id_tensor: |
| 173 | + merge_list = merge_list[:-1] |
| 174 | + phones_list = [] |
| 175 | + phones_list.append(merge_list) |
| 176 | + |
| 177 | + result["phone_ids"] = phones_list |
| 178 | + |
| 179 | + return result |
0 commit comments