Spaces:
Running
Running
File size: 5,347 Bytes
c282b65 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 |
import json
import os
from collections import defaultdict
from random import shuffle
from typing import Optional
import click
from tqdm import tqdm
from config import config
from text.cleaner import clean_text
from common.stdout_wrapper import SAFE_STDOUT
from common.log import logger
preprocess_text_config = config.preprocess_text_config
@click.command()
@click.option(
"--transcription-path",
default=preprocess_text_config.transcription_path,
type=click.Path(exists=True, file_okay=True, dir_okay=False),
)
@click.option("--cleaned-path", default=preprocess_text_config.cleaned_path)
@click.option("--train-path", default=preprocess_text_config.train_path)
@click.option("--val-path", default=preprocess_text_config.val_path)
@click.option(
"--config-path",
default=preprocess_text_config.config_path,
type=click.Path(exists=True, file_okay=True, dir_okay=False),
)
@click.option("--val-per-lang", default=preprocess_text_config.val_per_lang)
@click.option("--max-val-total", default=preprocess_text_config.max_val_total)
@click.option("--clean/--no-clean", default=preprocess_text_config.clean)
@click.option("-y", "--yml_config")
@click.option("--use_jp_extra", is_flag=True)
def preprocess(
transcription_path: str,
cleaned_path: Optional[str],
train_path: str,
val_path: str,
config_path: str,
val_per_lang: int,
max_val_total: int,
clean: bool,
yml_config: str, # 这个不要删
use_jp_extra: bool,
):
if cleaned_path == "" or cleaned_path is None:
cleaned_path = transcription_path + ".cleaned"
if clean:
with open(cleaned_path, "w", encoding="utf-8") as out_file:
with open(transcription_path, "r", encoding="utf-8") as trans_file:
for line in tqdm(trans_file, file=SAFE_STDOUT):
try:
utt, spk, language, text = line.strip().split("|")
norm_text, phones, tones, word2ph = clean_text(
text, language, use_jp_extra
)
out_file.write(
"{}|{}|{}|{}|{}|{}|{}\n".format(
utt,
spk,
language,
norm_text,
" ".join(phones),
" ".join([str(i) for i in tones]),
" ".join([str(i) for i in word2ph]),
)
)
except Exception as e:
logger.error(
f"An error occurred while generating the training set and validation set, at line:\n{line}\nDetails:\n{e}"
)
raise
transcription_path = cleaned_path
spk_utt_map = defaultdict(list)
spk_id_map = {}
current_sid = 0
with open(transcription_path, "r", encoding="utf-8") as f:
audioPaths = set()
countSame = 0
countNotFound = 0
for line in f.readlines():
utt, spk, language, text, phones, tones, word2ph = line.strip().split("|")
if utt in audioPaths:
# 过滤数据集错误：相同的音频匹配多个文本，导致后续bert出问题
logger.warning(f"Same audio matches multiple texts: {line}")
countSame += 1
continue
if not os.path.isfile(utt):
# 过滤数据集错误：不存在对应音频
logger.warning(f"Audio not found: {utt}")
countNotFound += 1
continue
audioPaths.add(utt)
spk_utt_map[language].append(line)
if spk not in spk_id_map.keys():
spk_id_map[spk] = current_sid
current_sid += 1
logger.info(
f"Total repeated audios: {countSame}, Total number of audio not found: {countNotFound}"
)
train_list = []
val_list = []
for spk, utts in spk_utt_map.items():
shuffle(utts)
val_list += utts[:val_per_lang]
train_list += utts[val_per_lang:]
shuffle(val_list)
if len(val_list) > max_val_total:
train_list += val_list[max_val_total:]
val_list = val_list[:max_val_total]
with open(train_path, "w", encoding="utf-8") as f:
for line in train_list:
f.write(line)
with open(val_path, "w", encoding="utf-8") as f:
for line in val_list:
f.write(line)
json_config = json.load(open(config_path, encoding="utf-8"))
json_config["data"]["spk2id"] = spk_id_map
json_config["data"]["n_speakers"] = len(spk_id_map)
# 新增写入：写入训练版本、数据集路径
# json_config["version"] = latest_version
json_config["data"]["training_files"] = os.path.normpath(train_path).replace(
"\\", "/"
)
json_config["data"]["validation_files"] = os.path.normpath(val_path).replace(
"\\", "/"
)
with open(config_path, "w", encoding="utf-8") as f:
json.dump(json_config, f, indent=2, ensure_ascii=False)
logger.info("Training set and validation set generation from texts is complete!")
if __name__ == "__main__":
preprocess()
|