tuna2134/sbv2-v2
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Fmt only (maybe)

Browse Source
This commit is contained in:
litagin02
2024-05-25 18:15:36 +09:00
parent acf93b80ce
commit 2274087da4
33 changed files with 216 additions and 166 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
app.py
View File

@@ -4,6 +4,7 @@ from pathlib import Path
import gradio as gr
import torch
from config import get_path_config
from gradio_tabs.dataset import create_dataset_app
from gradio_tabs.inference import create_inference_app
from gradio_tabs.merge import create_merge_app
@@ -13,7 +14,6 @@ from style_bert_vits2.constants import GRADIO_THEME, VERSION
from style_bert_vits2.nlp.japanese import pyopenjtalk_worker
from style_bert_vits2.nlp.japanese.user_dict import update_dict
from style_bert_vits2.tts_model import TTSModelHolder
from config import get_path_config
# このプロセスからはワーカーを起動して辞書を使いたいので、ここで初期化

9
bert_gen.py
View File

@@ -10,10 +10,7 @@ from style_bert_vits2.constants import Languages
from style_bert_vits2.logging import logger
from style_bert_vits2.models import commons
from style_bert_vits2.models.hyper_parameters import HyperParameters
from style_bert_vits2.nlp import (
cleaned_text_to_sequence,
extract_bert_feature,
)
from style_bert_vits2.nlp import cleaned_text_to_sequence, extract_bert_feature
from style_bert_vits2.nlp.japanese import pyopenjtalk_worker
from style_bert_vits2.nlp.japanese.user_dict import update_dict
from style_bert_vits2.utils.stdout_wrapper import SAFE_STDOUT
@@ -77,10 +74,10 @@ if __name__ == "__main__":
config_path = args.config
hps = HyperParameters.load_from_json(config_path)
lines: list[str] = []
with open(hps.data.training_files, "r", encoding="utf-8") as f:
with open(hps.data.training_files, encoding="utf-8") as f:
lines.extend(f.readlines())
with open(hps.data.validation_files, "r", encoding="utf-8") as f:
with open(hps.data.validation_files, encoding="utf-8") as f:
lines.extend(f.readlines())
add_blank = [hps.data.add_blank] * len(lines)

14
config.py
View File

@@ -56,8 +56,13 @@ class Preprocess_text_config:
clean: bool = True,
):
self.transcription_path = Path(transcription_path)
self.cleaned_path = Path(cleaned_path)
self.train_path = Path(train_path)
if cleaned_path == "" or cleaned_path is None:
self.cleaned_path = self.transcription_path.with_name(
self.transcription_path.name + ".cleaned"
)
else:
self.cleaned_path = Path(cleaned_path)
self.val_path = Path(val_path)
self.config_path = Path(config_path)
self.val_per_lang = val_per_lang
@@ -70,7 +75,7 @@ class Preprocess_text_config:
data["transcription_path"] = dataset_path / data["transcription_path"]
if data["cleaned_path"] == "" or data["cleaned_path"] is None:
data["cleaned_path"] = None
data["cleaned_path"] = ""
else:
data["cleaned_path"] = dataset_path / data["cleaned_path"]
data["train_path"] = dataset_path / data["train_path"]
@@ -232,7 +237,7 @@ class Config:
"Please do not modify default_config.yml. Instead, modify config.yml."
)
# sys.exit(0)
with open(config_path, "r", encoding="utf-8") as file:
with open(config_path, encoding="utf-8") as file:
yaml_config: dict[str, Any] = yaml.safe_load(file.read())
model_name: str = yaml_config["model_name"]
self.model_name: str = model_name
@@ -241,6 +246,7 @@ class Config:
else:
dataset_path = path_config.dataset_root / model_name
self.dataset_path = dataset_path
self.dataset_root = path_config.dataset_root
self.assets_root = path_config.assets_root
self.out_dir = self.assets_root / model_name
self.resample_config: Resample_config = Resample_config.from_dict(
@@ -284,7 +290,7 @@ def get_path_config() -> PathConfig:
logger.info(
"Please do not modify configs/default_paths.yml. Instead, modify configs/paths.yml."
)
with open(path_config_path, "r", encoding="utf-8") as file:
with open(path_config_path, encoding="utf-8") as file:
path_config_dict: dict[str, str] = yaml.safe_load(file.read())
return PathConfig(**path_config_dict)

7
data_utils.py
View File

@@ -7,7 +7,7 @@ import torch
import torch.utils.data
from tqdm import tqdm
from config import config
from config import get_config
from mel_processing import mel_spectrogram_torch, spectrogram_torch
from style_bert_vits2.logging import logger
from style_bert_vits2.models import commons
@@ -16,6 +16,7 @@ from style_bert_vits2.models.utils import load_filepaths_and_text, load_wav_to_t
from style_bert_vits2.nlp import cleaned_text_to_sequence
config = get_config()
"""Multi speaker version"""
@@ -120,9 +121,7 @@ class TextAudioSpeakerLoader(torch.utils.data.Dataset):
audio, sampling_rate = load_wav_to_torch(filename)
if sampling_rate != self.sampling_rate:
raise ValueError(
"{} {} SR doesn't match target {} SR".format(
filename, sampling_rate, self.sampling_rate
)
f"{filename} {sampling_rate} SR doesn't match target {self.sampling_rate} SR"
)
audio_norm = audio / self.max_wav_value
audio_norm = audio_norm.unsqueeze(0)

6
default_style.py
View File

@@ -33,7 +33,7 @@ def save_neutral_vector(wav_dir: Union[Path, str], output_path: Union[Path, str]
np.save(output_path, only_mean)
logger.info(f"Saved mean style vector to {output_path}")
with open(json_path, "r", encoding="utf-8") as f:
with open(json_path, encoding="utf-8") as f:
json_dict = json.load(f)
json_dict["data"]["num_styles"] = 1
json_dict["data"]["style2id"] = {DEFAULT_STYLE: 0}
@@ -50,7 +50,7 @@ def save_styles_by_dirs(wav_dir: Union[Path, str], output_dir: Union[Path, str])
subdirs = [d for d in wav_dir.iterdir() if d.is_dir()]
subdirs.sort()
if len(subdirs) in (0, 1):
if not subdirs:
logger.warning("No style directories found. Saving only neutral style.")
save_neutral_vector(wav_dir, output_dir)
@@ -85,7 +85,7 @@ def save_styles_by_dirs(wav_dir: Union[Path, str], output_dir: Union[Path, str])
# Save style2id config to json
style2id = {name: i for i, name in enumerate(names)}
with open(json_path, "r", encoding="utf-8") as f:
with open(json_path, encoding="utf-8") as f:
json_dict = json.load(f)
json_dict["data"]["num_styles"] = len(names)
json_dict["data"]["style2id"] = style2id

2
gen_yaml.py
View File

@@ -22,7 +22,7 @@ args = parser.parse_args()
def gen_yaml(model_name, dataset_path):
if not os.path.exists("config.yml"):
shutil.copy(src="default_config.yml", dst="config.yml")
with open("config.yml", "r", encoding="utf-8") as f:
with open("config.yml", encoding="utf-8") as f:
yml_data = yaml.safe_load(f)
yml_data["model_name"] = model_name
yml_data["dataset_path"] = dataset_path

12
gradio_tabs/merge.py
View File

@@ -47,9 +47,9 @@ def merge_style(
style_vectors_b = np.load(
assets_root / model_name_b / "style_vectors.npy"
) # (style_num_b, 256)
with open(assets_root / model_name_a / "config.json", "r", encoding="utf-8") as f:
with open(assets_root / model_name_a / "config.json", encoding="utf-8") as f:
config_a = json.load(f)
with open(assets_root / model_name_b / "config.json", "r", encoding="utf-8") as f:
with open(assets_root / model_name_b / "config.json", encoding="utf-8") as f:
config_b = json.load(f)
style2id_a = config_a["data"]["style2id"]
style2id_b = config_b["data"]["style2id"]
@@ -83,7 +83,7 @@ def merge_style(
# recipe.jsonを読み込んで、style_triple_listを追記
info_path = assets_root / output_name / "recipe.json"
if info_path.exists():
with open(info_path, "r", encoding="utf-8") as f:
with open(info_path, encoding="utf-8") as f:
info = json.load(f)
else:
info = {}
@@ -143,7 +143,7 @@ def merge_models(
merged_model_weight = model_a_weight.copy()
for key in model_a_weight.keys():
for key in model_a_weight:
if any([key.startswith(prefix) for prefix in voice_keys]):
weight = voice_weight
elif any([key.startswith(prefix) for prefix in voice_pitch_keys]):
@@ -256,12 +256,12 @@ def update_two_model_names_dropdown(model_holder: TTSModelHolder):
def load_styles_gr(model_name_a: str, model_name_b: str):
config_path_a = assets_root / model_name_a / "config.json"
with open(config_path_a, "r", encoding="utf-8") as f:
with open(config_path_a, encoding="utf-8") as f:
config_a = json.load(f)
styles_a = list(config_a["data"]["style2id"].keys())
config_path_b = assets_root / model_name_b / "config.json"
with open(config_path_b, "r", encoding="utf-8") as f:
with open(config_path_b, encoding="utf-8") as f:
config_b = json.load(f)
styles_b = list(config_b["data"]["style2id"].keys())

9
gradio_tabs/train.py
View File

@@ -5,13 +5,14 @@ import subprocess
import sys
import time
import webbrowser
from dataclasses import dataclass
from datetime import datetime
from multiprocessing import cpu_count
from pathlib import Path
import gradio as gr
import yaml
from dataclasses import dataclass
from config import get_path_config
from style_bert_vits2.logging import logger
from style_bert_vits2.utils.stdout_wrapper import SAFE_STDOUT
@@ -75,7 +76,7 @@ def initialize(
"configs/config.json" if not use_jp_extra else "configs/config_jp_extra.json"
)
with open(default_config_path, "r", encoding="utf-8") as f:
with open(default_config_path, encoding="utf-8") as f:
config = json.load(f)
config["model_name"] = model_name
config["data"]["training_files"] = str(paths.train_path)
@@ -121,7 +122,7 @@ def initialize(
json.dump(config, f, indent=2, ensure_ascii=False)
if not Path("config.yml").exists():
shutil.copy(src="default_config.yml", dst="config.yml")
with open("config.yml", "r", encoding="utf-8") as f:
with open("config.yml", encoding="utf-8") as f:
yml_data = yaml.safe_load(f)
yml_data["model_name"] = model_name
yml_data["dataset_path"] = str(paths.dataset_path)
@@ -331,7 +332,7 @@ def train(
):
paths = get_path(model_name)
# 学習再開の場合を考えて念のためconfig.ymlの名前等を更新
with open("config.yml", "r", encoding="utf-8") as f:
with open("config.yml", encoding="utf-8") as f:
yml_data = yaml.safe_load(f)
yml_data["model_name"] = model_name
yml_data["dataset_path"] = str(paths.dataset_path)

4
initialize.py
View File

@@ -10,7 +10,7 @@ from style_bert_vits2.logging import logger
def download_bert_models():
with open("bert/bert_models.json", "r", encoding="utf-8") as fp:
with open("bert/bert_models.json", encoding="utf-8") as fp:
models = json.load(fp)
for k, v in models.items():
local_path = Path("bert").joinpath(k)
@@ -113,7 +113,7 @@ def main():
return
# Change default paths if necessary
with open(paths_yml, "r", encoding="utf-8") as f:
with open(paths_yml, encoding="utf-8") as f:
yml_data = yaml.safe_load(f)
if args.assets_root is not None:
yml_data["assets_root"] = args.assets_root

2
preprocess_text.py
View File

@@ -145,7 +145,7 @@ def preprocess(
spk_utt_map[spk].append(line)
# 新しい話者が出てきたら話者IDを割り当て、current_sidを1増やす
if spk not in spk_id_map.keys():
if spk not in spk_id_map:
spk_id_map[spk] = current_sid
current_sid += 1
if count_same > 0 or count_not_found > 0:

74
pyproject.toml
View File

@@ -5,7 +5,7 @@ build-backend = "hatchling.build"
[project]
name = "style-bert-vits2"
dynamic = ["version"]
description = 'Style-Bert-VITS2: Bert-VITS2 with more controllable voice styles.'
description = "Style-Bert-VITS2: Bert-VITS2 with more controllable voice styles."
readme = "README.md"
requires-python = ">=3.9"
license = "AGPL-3.0"
@@ -22,21 +22,21 @@ classifiers = [
"Programming Language :: Python :: Implementation :: CPython",
]
dependencies = [
'cmudict',
'cn2an',
'g2p_en',
'jieba',
'loguru',
'num2words',
'numba',
'numpy',
'pydantic>=2.0',
'pyopenjtalk-dict',
'pypinyin',
'pyworld-prebuilt',
'safetensors',
'torch>=2.1',
'transformers',
"cmudict",
"cn2an",
"g2p_en",
"jieba",
"loguru",
"num2words",
"numba",
"numpy",
"pydantic>=2.0",
"pyopenjtalk-dict",
"pypinyin",
"pyworld-prebuilt",
"safetensors",
"torch>=2.1",
"transformers",
]
[project.urls]
@@ -59,42 +59,26 @@ only-include = [
"pyproject.toml",
"README.md",
]
exclude = [
".git",
".gitignore",
".gitattributes",
]
exclude = [".git", ".gitignore", ".gitattributes"]
[tool.hatch.build.targets.wheel]
packages = ["style_bert_vits2"]
[tool.hatch.envs.test]
dependencies = [
"coverage[toml]>=6.5",
"pytest",
]
dependencies = ["coverage[toml]>=6.5", "pytest"]
[tool.hatch.envs.test.scripts]
# Usage: `hatch run test:test`
test = "pytest {args:tests}"
# Usage: `hatch run test:coverage`
test-cov = "coverage run -m pytest {args:tests}"
# Usage: `hatch run test:cov-report`
cov-report = [
"- coverage combine",
"coverage report",
]
cov-report = ["- coverage combine", "coverage report"]
# Usage: `hatch run test:cov`
cov = [
"test-cov",
"cov-report",
]
cov = ["test-cov", "cov-report"]
[tool.hatch.envs.style]
detached = true
dependencies = [
"black",
"isort",
]
dependencies = ["black", "isort"]
[tool.hatch.envs.style.scripts]
check = [
"black --check --diff .",
@@ -113,17 +97,17 @@ python = ["3.9", "3.10", "3.11"]
source_pkgs = ["style_bert_vits2", "tests"]
branch = true
parallel = true
omit = [
"style_bert_vits2/constants.py",
]
omit = ["style_bert_vits2/constants.py"]
[tool.coverage.paths]
style_bert_vits2 = ["style_bert_vits2", "*/style-bert-vits2/style_bert_vits2"]
tests = ["tests", "*/style-bert-vits2/tests"]
[tool.coverage.report]
exclude_lines = [
"no cov",
"if __name__ == .__main__.:",
"if TYPE_CHECKING:",
]
exclude_lines = ["no cov", "if __name__ == .__main__.:", "if TYPE_CHECKING:"]
[tool.ruff]
extend-select = ["I"]
[tool.ruff.lint.isort]
lines-after-imports = 2

2
server_editor.py
View File

@@ -127,7 +127,7 @@ def download_and_extract(url, extract_to: Path):
def new_release_available(latest_release):
if LAST_DOWNLOAD_FILE.exists():
with open(LAST_DOWNLOAD_FILE, "r") as file:
with open(LAST_DOWNLOAD_FILE) as file:
last_download_str = file.read().strip()
# 'Z'を除去して日時オブジェクトに変換
last_download_str = last_download_str.replace("Z", "+00:00")

1
slice.py
View File

@@ -7,7 +7,6 @@ from typing import Any, Optional
import soundfile as sf
import torch
import yaml
from tqdm import tqdm
from config import get_path_config

2
style_bert_vits2/models/hyper_parameters.py
View File

@@ -125,5 +125,5 @@ class HyperParameters(BaseModel):
HyperParameters: ハイパーパラメータ
"""
with open(json_path, "r", encoding="utf-8") as f:
with open(json_path, encoding="utf-8") as f:
return HyperParameters.model_validate_json(f.read())

2
style_bert_vits2/models/models.py
View File

@@ -786,7 +786,7 @@ class ReferenceEncoder(nn.Module):
for i in range(K)
]
self.convs = nn.ModuleList(convs)
# self.wns = nn.ModuleList([weight_norm(num_features=ref_enc_filters[i]) for i in range(K)]) # noqa: E501
# self.wns = nn.ModuleList([weight_norm(num_features=ref_enc_filters[i]) for i in range(K)])
out_channels = self.calculate_channels(spec_channels, 3, 2, 1, K)
self.gru = nn.GRU(

2
style_bert_vits2/models/models_jp_extra.py
View File

@@ -844,7 +844,7 @@ class ReferenceEncoder(nn.Module):
for i in range(K)
]
self.convs = nn.ModuleList(convs)
# self.wns = nn.ModuleList([weight_norm(num_features=ref_enc_filters[i]) for i in range(K)]) # noqa: E501
# self.wns = nn.ModuleList([weight_norm(num_features=ref_enc_filters[i]) for i in range(K)])
out_channels = self.calculate_channels(spec_channels, 3, 2, 1, K)
self.gru = nn.GRU(

12
style_bert_vits2/models/utils/__init__.py
View File

@@ -186,7 +186,7 @@ def load_filepaths_and_text(
list[list[str]]: ファイルパスとテキストのリスト
"""
with open(filename, "r", encoding="utf-8") as f:
with open(filename, encoding="utf-8") as f:
filepaths_and_text = [line.strip().split(split) for line in f]
return filepaths_and_text
@@ -245,9 +245,7 @@ def check_git_hash(model_dir_path: Union[str, Path]) -> None:
source_dir = os.path.dirname(os.path.realpath(__file__))
if not os.path.exists(os.path.join(source_dir, ".git")):
logger.warning(
"{} is not a git repository, therefore hash value comparison will be ignored.".format(
source_dir
)
f"{source_dir} is not a git repository, therefore hash value comparison will be ignored."
)
return
@@ -255,13 +253,11 @@ def check_git_hash(model_dir_path: Union[str, Path]) -> None:
path = os.path.join(model_dir_path, "githash")
if os.path.exists(path):
with open(path, "r", encoding="utf-8") as f:
with open(path, encoding="utf-8") as f:
saved_hash = f.read()
if saved_hash != cur_hash:
logger.warning(
"git hash values are different. {}(saved) != {}(current)".format(
saved_hash[:8], cur_hash[:8]
)
f"git hash values are different. {saved_hash[:8]}(saved) != {cur_hash[:8]}(current)"
)
else:
with open(path, "w", encoding="utf-8") as f:

2
style_bert_vits2/models/utils/safetensors.py
View File

@@ -77,7 +77,7 @@ def save_safetensors(
keys = []
for k in state_dict:
if "enc_q" in k and for_infer:
continue # noqa: E701
continue
keys.append(k)
new_dict = (

10
style_bert_vits2/nlp/chinese/g2p.py
View File

@@ -8,7 +8,7 @@ from style_bert_vits2.nlp.chinese.tone_sandhi import ToneSandhi
from style_bert_vits2.nlp.symbols import PUNCTUATIONS
with open(Path(__file__).parent / "opencpop-strict.txt", "r", encoding="utf-8") as f:
with open(Path(__file__).parent / "opencpop-strict.txt", encoding="utf-8") as f:
__PINYIN_TO_SYMBOL_MAP = {
line.split("\t")[0]: line.strip().split("\t")[1] for line in f.readlines()
}
@@ -73,7 +73,7 @@ def __g2p(segments: list[str]) -> tuple[list[str], list[int], list[int]]:
"iou": "iu",
"uen": "un",
}
if v_without_tone in v_rep_map.keys():
if v_without_tone in v_rep_map:
pinyin = c + v_rep_map[v_without_tone]
else:
# 单音节
@@ -83,7 +83,7 @@ def __g2p(segments: list[str]) -> tuple[list[str], list[int], list[int]]:
"in": "yin",
"u": "wu",
}
if pinyin in pinyin_rep_map.keys():
if pinyin in pinyin_rep_map:
pinyin = pinyin_rep_map[pinyin]
else:
single_rep_map = {
@@ -92,10 +92,10 @@ def __g2p(segments: list[str]) -> tuple[list[str], list[int], list[int]]:
"i": "y",
"u": "w",
}
if pinyin[0] in single_rep_map.keys():
if pinyin[0] in single_rep_map:
pinyin = single_rep_map[pinyin[0]] + pinyin[1:]
assert pinyin in __PINYIN_TO_SYMBOL_MAP.keys(), (
assert pinyin in __PINYIN_TO_SYMBOL_MAP, (
pinyin,
seg,
raw_pinyin,

2
style_bert_vits2/nlp/chinese/normalizer.py
View File

@@ -51,7 +51,7 @@ def normalize_text(text: str) -> str:
def replace_punctuation(text: str) -> str:
text = text.replace("嗯", "恩").replace("呣", "母")
pattern = re.compile("|".join(re.escape(p) for p in __REPLACE_MAP.keys()))
pattern = re.compile("|".join(re.escape(p) for p in __REPLACE_MAP))
replaced_text = pattern.sub(lambda x: __REPLACE_MAP[x.group()], text)

54
style_bert_vits2/nlp/chinese/tone_sandhi.py
View File

@@ -471,26 +471,27 @@ class ToneSandhi:
):
finals[j] = finals[j][:-1] + "5"
ge_idx = word.find("个")
if len(word) >= 1 and word[-1] in "吧呢啊呐噻嘛吖嗨呐哦哒额滴哩哟喽啰耶喔诶":
finals[-1] = finals[-1][:-1] + "5"
elif len(word) >= 1 and word[-1] in "的地得":
finals[-1] = finals[-1][:-1] + "5"
# e.g. 走了, 看着, 去过
# elif len(word) == 1 and word in "了着过" and pos in {"ul", "uz", "ug"}:
# finals[-1] = finals[-1][:-1] + "5"
elif (
len(word) > 1
and word[-1] in "们子"
and pos in {"r", "n"}
and word not in self.must_not_neural_tone_words
if (
len(word) >= 1
and word[-1] in "吧呢啊呐噻嘛吖嗨呐哦哒额滴哩哟喽啰耶喔诶"
or len(word) >= 1
and word[-1] in "的地得"
or (
(
len(word) > 1
and word[-1] in "们子"
and pos in {"r", "n"}
and word not in self.must_not_neural_tone_words
)
or len(word) > 1
and word[-1] in "上下里"
and pos in {"s", "l", "f"}
)
or len(word) > 1
and word[-1] in "来去"
and word[-2] in "上下进出回过起开"
):
finals[-1] = finals[-1][:-1] + "5"
# e.g. 桌上, 地下, 家里
elif len(word) > 1 and word[-1] in "上下里" and pos in {"s", "l", "f"}:
finals[-1] = finals[-1][:-1] + "5"
# e.g. 上来, 下去
elif len(word) > 1 and word[-1] in "来去" and word[-2] in "上下进出回过起开":
finals[-1] = finals[-1][:-1] + "5"
# 个做量词
elif (
ge_idx >= 1
@@ -500,12 +501,11 @@ class ToneSandhi:
)
) or word == "个":
finals[ge_idx] = finals[ge_idx][:-1] + "5"
else:
if (
word in self.must_neural_tone_words
or word[-2:] in self.must_neural_tone_words
):
finals[-1] = finals[-1][:-1] + "5"
elif (
word in self.must_neural_tone_words
or word[-2:] in self.must_neural_tone_words
):
finals[-1] = finals[-1][:-1] + "5"
word_list = self._split_word(word)
finals_list = [finals[: len(word_list[0])], finals[len(word_list[0]) :]]
@@ -549,10 +549,8 @@ class ToneSandhi:
if finals[i + 1][-1] == "4":
finals[i] = finals[i][:-1] + "2"
# "一" before non-tone4 should be yi4, e.g. 一天
else:
# "一" 后面如果是标点，还读一声
if word[i + 1] not in self.punc:
finals[i] = finals[i][:-1] + "4"
elif word[i + 1] not in self.punc:
finals[i] = finals[i][:-1] + "4"
return finals
def _split_word(self, word: str) -> list[str]:

2
style_bert_vits2/nlp/english/cmudict.py
View File

@@ -20,7 +20,7 @@ def get_dict() -> dict[str, list[list[str]]]:
def read_dict() -> dict[str, list[list[str]]]:
g2p_dict = {}
start_line = 49
with open(CMU_DICT_PATH, "r", encoding="utf-8") as f:
with open(CMU_DICT_PATH, encoding="utf-8") as f:
line = f.readline()
line_index = 1
while line:

126
style_bert_vits2/nlp/english/g2p.py
View File

@@ -1,23 +1,91 @@
import re
from g2p_en import G2p
from style_bert_vits2.constants import Languages
from style_bert_vits2.nlp import bert_models
from style_bert_vits2.nlp.english.cmudict import get_dict
from style_bert_vits2.nlp.symbols import PUNCTUATIONS, SYMBOLS
# Initialize global variables once
ARPA = {
"AH0", "S", "AH1", "EY2", "AE2", "EH0", "OW2", "UH0", "NG", "B", "G", "AY0",
"M", "AA0", "F", "AO0", "ER2", "UH1", "IY1", "AH2", "DH", "IY0", "EY1",
"IH0", "K", "N", "W", "IY2", "T", "AA1", "ER1", "EH2", "OY0", "UH2", "UW1",
"Z", "AW2", "AW1", "V", "UW2", "AA2", "ER", "AW0", "UW0", "R", "OW1", "EH1",
"ZH", "AE0", "IH2", "IH", "Y", "JH", "P", "AY1", "EY0", "OY2", "TH", "HH",
"D", "ER0", "CH", "AO1", "AE1", "AO2", "OY1", "AY2", "IH1", "OW0", "L",
"SH"
"AH0",
"S",
"AH1",
"EY2",
"AE2",
"EH0",
"OW2",
"UH0",
"NG",
"B",
"G",
"AY0",
"M",
"AA0",
"F",
"AO0",
"ER2",
"UH1",
"IY1",
"AH2",
"DH",
"IY0",
"EY1",
"IH0",
"K",
"N",
"W",
"IY2",
"T",
"AA1",
"ER1",
"EH2",
"OY0",
"UH2",
"UW1",
"Z",
"AW2",
"AW1",
"V",
"UW2",
"AA2",
"ER",
"AW0",
"UW0",
"R",
"OW1",
"EH1",
"ZH",
"AE0",
"IH2",
"IH",
"Y",
"JH",
"P",
"AY1",
"EY0",
"OY2",
"TH",
"HH",
"D",
"ER0",
"CH",
"AO1",
"AE1",
"AO2",
"OY1",
"AY2",
"IH1",
"OW0",
"L",
"SH",
}
_g2p = G2p()
eng_dict = get_dict()
def g2p(text: str) -> tuple[list[str], list[int], list[int]]:
phones = []
tones = []
@@ -51,7 +119,7 @@ def g2p(text: str) -> tuple[list[str], list[int], list[int]]:
tns.append(0)
temp_phones += [__post_replace_ph(i) for i in phns]
temp_tones += tns
phones += temp_phones
tones += temp_tones
phone_len.append(len(temp_phones))
@@ -72,9 +140,19 @@ def g2p(text: str) -> tuple[list[str], list[int], list[int]]:
def __post_replace_ph(ph: str) -> str:
REPLACE_MAP = {
"：": ",", "；": ",", "，": ",", "。": ".", "！": "!", "？": "?",
"\n": ".", "·": ",", "、": ",", "…": "...", "···": "...",
"・・・": "...", "v": "V"
"：": ",",
"；": ",",
"，": ",",
"。": ".",
"！": "!",
"？": "?",
"\n": ".",
"·": ",",
"、": ",",
"…": "...",
"···": "...",
"・・・": "...",
"v": "V",
}
if ph in REPLACE_MAP:
ph = REPLACE_MAP[ph]
@@ -120,21 +198,22 @@ def __text_to_words(text: str) -> list[list[str]]:
for idx, t in enumerate(tokens):
if t.startswith("▁"):
words.append([t[1:]])
else:
if t in PUNCTUATIONS:
if idx == len(tokens) - 1:
words.append([f"{t}"])
else:
if not tokens[idx + 1].startswith("▁") and tokens[idx + 1] not in PUNCTUATIONS:
if idx == 0:
words.append([])
words[-1].append(f"{t}")
else:
words.append([f"{t}"])
else:
elif t in PUNCTUATIONS:
if idx == len(tokens) - 1:
words.append([f"{t}"])
elif (
not tokens[idx + 1].startswith("▁")
and tokens[idx + 1] not in PUNCTUATIONS
):
if idx == 0:
words.append([])
words[-1].append(f"{t}")
else:
words.append([f"{t}"])
else:
if idx == 0:
words.append([])
words[-1].append(f"{t}")
return words
@@ -149,4 +228,3 @@ if __name__ == "__main__":
# for ph in group:
# all_phones.add(ph)
# print(all_phones)

2
style_bert_vits2/nlp/english/normalizer.py
View File

@@ -58,7 +58,7 @@ def replace_punctuation(text: str) -> str:
"「": "'",
"」": "'",
}
pattern = re.compile("|".join(re.escape(p) for p in REPLACE_MAP.keys()))
pattern = re.compile("|".join(re.escape(p) for p in REPLACE_MAP))
replaced_text = pattern.sub(lambda x: REPLACE_MAP[x.group()], text)
# replaced_text = re.sub(
# r"[^\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3400-\u4DBF\u3005"

2
style_bert_vits2/nlp/japanese/g2p.py
View File

@@ -719,5 +719,3 @@ class YomiError(Exception):
基本的に「学習の前処理のテキスト処理時」には発生させ、そうでない場合は、
ignore_yomi_error=True にしておいて、この例外を発生させないようにする。
"""
pass

2
style_bert_vits2/nlp/japanese/normalizer.py
View File

@@ -60,7 +60,7 @@ __REPLACE_MAP = {
"」": "'",
}
# 記号類の正規化パターン
__REPLACE_PATTERN = re.compile("|".join(re.escape(p) for p in __REPLACE_MAP.keys()))
__REPLACE_PATTERN = re.compile("|".join(re.escape(p) for p in __REPLACE_MAP))
# 句読点等の正規化パターン
__PUNCTUATION_CLEANUP_PATTERN = re.compile(
# ↓ ひらがな、カタカナ、漢字

4
style_bert_vits2/nlp/japanese/pyopenjtalk_worker/__init__.py
View File

@@ -88,7 +88,7 @@ def initialize_worker(port: int = WORKER_PORT) -> None:
client = None
try:
client = WorkerClient(port)
except (socket.timeout, socket.error):
except (OSError, socket.timeout):
logger.debug("try starting pyopenjtalk worker server")
import os
import subprocess
@@ -120,7 +120,7 @@ def initialize_worker(port: int = WORKER_PORT) -> None:
try:
client = WorkerClient(port)
break
except socket.error:
except OSError:
time.sleep(0.5)
count += 1
# 20: max number of retries

2
style_bert_vits2/nlp/japanese/user_dict/word_model.py
View File

@@ -114,7 +114,7 @@ class PartOfSpeechDetail(BaseModel):
part_of_speech_detail_2: str = Field(title="品詞細分類2")
part_of_speech_detail_3: str = Field(title="品詞細分類3")
# context_idは辞書の左・右文脈IDのこと
# https://github.com/VOICEVOX/open_jtalk/blob/427cfd761b78efb6094bea3c5bb8c968f0d711ab/src/mecab-naist-jdic/_left-id.def # noqa
# https://github.com/VOICEVOX/open_jtalk/blob/427cfd761b78efb6094bea3c5bb8c968f0d711ab/src/mecab-naist-jdic/_left-id.def
context_id: int = Field(title="文脈ID")
cost_candidates: List[int] = Field(title="コストのパーセンタイル")
accent_associative_rules: List[str] = Field(title="アクセント結合規則の一覧")

2
style_bert_vits2/tts_model.py
View File

@@ -1,5 +1,5 @@
from pathlib import Path
from typing import TYPE_CHECKING, Any, Optional, Union
from typing import Any, Optional, Union
import numpy as np
import torch

1
style_bert_vits2/utils/subprocess.py
View File

@@ -27,6 +27,7 @@ def run_script_with_log(
stderr=subprocess.PIPE,
text=True,
encoding="utf-8",
check=False,
)
if result.returncode != 0:
logger.error(f"Error: {' '.join(cmd)}\n{result.stderr}")

1
style_gen.py
View File

@@ -26,7 +26,6 @@ class NaNValueError(ValueError):
"""カスタム例外クラス。NaN値が見つかった場合に使用されます。"""
# 推論時にインポートするために短いが関数を書く
def get_style_vector(wav_path: str) -> NDArray[Any]:
return inference(wav_path) # type: ignore

6
train_ms_jp_extra.py
View File

@@ -17,11 +17,7 @@ from tqdm import tqdm
# logging.getLogger("numba").setLevel(logging.WARNING)
import default_style
from config import get_config
from data_utils import (
DistributedBucketSampler,
TextAudioSpeakerCollate,
TextAudioSpeakerLoader,
)
from data_utils import TextAudioSpeakerCollate, TextAudioSpeakerLoader
from losses import WavLMLoss, discriminator_loss, feature_loss, generator_loss, kl_loss
from mel_processing import mel_spectrogram_torch, spec_to_mel_torch
from style_bert_vits2.logging import logger

2
transcribe.py
View File

@@ -1,10 +1,8 @@
import argparse
import os
import sys
from pathlib import Path
from typing import Any, Optional
import yaml
from torch.utils.data import Dataset
from tqdm import tqdm