chenjgtea
commited on
Commit
•
f097912
1
Parent(s):
52b0147
gpu模式下新增中英文的normal花处理
Browse files- Chat2TTS/core.py +31 -3
- requirements.txt +4 -0
- tool/normalizer/__init__.py +1 -0
- tool/normalizer/normali_util.py +17 -0
- web/app_gpu.py +15 -4
Chat2TTS/core.py
CHANGED
@@ -13,6 +13,7 @@ from .infer.api import refine_text, infer_code
|
|
13 |
from dataclasses import dataclass
|
14 |
from typing import Literal, Optional, List, Tuple, Dict
|
15 |
from tool.logger import get_logger
|
|
|
16 |
|
17 |
from ChatTTS.norm import Normalizer
|
18 |
|
@@ -60,10 +61,36 @@ class Chat:
|
|
60 |
else:
|
61 |
self.logger.log(logging.INFO, f'Load from cache: {download_path}')
|
62 |
self._load(**{k: os.path.join(download_path, v) for k, v in OmegaConf.load(os.path.join(download_path, 'config', 'path.yaml')).items()})
|
|
|
63 |
elif source == 'local':
|
64 |
self.logger.log(logging.INFO, f'Load from local: {local_path}')
|
65 |
self._load(**{k: os.path.join(local_path, v) for k, v in OmegaConf.load(os.path.join(local_path, 'config', 'path.yaml')).items()})
|
66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
def _load(
|
68 |
self,
|
69 |
vocos_config_path: str = None,
|
@@ -156,7 +183,8 @@ class Chat:
|
|
156 |
refine_text_only=False,
|
157 |
params_refine_text={},
|
158 |
params_infer_code={},
|
159 |
-
use_decoder=False
|
|
|
160 |
):
|
161 |
|
162 |
assert self.check_model(use_decoder=use_decoder)
|
@@ -169,7 +197,7 @@ class Chat:
|
|
169 |
text=t,
|
170 |
do_text_normalization=True,
|
171 |
do_homophone_replacement=True,
|
172 |
-
lang=
|
173 |
)
|
174 |
for t in text
|
175 |
]
|
|
|
13 |
from dataclasses import dataclass
|
14 |
from typing import Literal, Optional, List, Tuple, Dict
|
15 |
from tool.logger import get_logger
|
16 |
+
from tool.normalizer import normalizer_en_nemo_text,normalizer_cn_tn
|
17 |
|
18 |
from ChatTTS.norm import Normalizer
|
19 |
|
|
|
61 |
else:
|
62 |
self.logger.log(logging.INFO, f'Load from cache: {download_path}')
|
63 |
self._load(**{k: os.path.join(download_path, v) for k, v in OmegaConf.load(os.path.join(download_path, 'config', 'path.yaml')).items()})
|
64 |
+
self._regist_normalizer()
|
65 |
elif source == 'local':
|
66 |
self.logger.log(logging.INFO, f'Load from local: {local_path}')
|
67 |
self._load(**{k: os.path.join(local_path, v) for k, v in OmegaConf.load(os.path.join(local_path, 'config', 'path.yaml')).items()})
|
68 |
+
|
69 |
+
def _regist_normalizer(self):
|
70 |
+
|
71 |
+
self.logger.info("==========开始注册 normalizer===========")
|
72 |
+
|
73 |
+
try:
|
74 |
+
self.normalizer.register("en",normalizer_en_nemo_text())
|
75 |
+
except ValueError as e:
|
76 |
+
self.logger.error('normalizer_en_nemo_text register fail' , e)
|
77 |
+
except:
|
78 |
+
self.logger.error("Package nemo_text_processing not found!")
|
79 |
+
self.logger.error(
|
80 |
+
"Run: conda install -c conda-forge pynini=2.1.5 && pip install nemo_text_processing",
|
81 |
+
)
|
82 |
+
|
83 |
+
try:
|
84 |
+
self.normalizer.register("zh",normalizer_cn_tn())
|
85 |
+
except ValueError as e:
|
86 |
+
self.logger.error('normalizer_cn_tn register fail' , e)
|
87 |
+
except:
|
88 |
+
self.logger.error("Package WeTextProcessing not found!")
|
89 |
+
self.logger.error(
|
90 |
+
"Run: conda install -c conda-forge pynini=2.1.5 && pip install WeTextProcessing",
|
91 |
+
)
|
92 |
+
|
93 |
+
|
94 |
def _load(
|
95 |
self,
|
96 |
vocos_config_path: str = None,
|
|
|
183 |
refine_text_only=False,
|
184 |
params_refine_text={},
|
185 |
params_infer_code={},
|
186 |
+
use_decoder=False,
|
187 |
+
lang=None
|
188 |
):
|
189 |
|
190 |
assert self.check_model(use_decoder=use_decoder)
|
|
|
197 |
text=t,
|
198 |
do_text_normalization=True,
|
199 |
do_homophone_replacement=True,
|
200 |
+
lang=lang,
|
201 |
)
|
202 |
for t in text
|
203 |
]
|
requirements.txt
CHANGED
@@ -20,6 +20,10 @@ vector_quantize_pytorch
|
|
20 |
# Hugging Face Hub client
|
21 |
huggingface_hub
|
22 |
|
|
|
|
|
|
|
|
|
23 |
vocos
|
24 |
|
25 |
spaces
|
|
|
20 |
# Hugging Face Hub client
|
21 |
huggingface_hub
|
22 |
|
23 |
+
pynini==2.1.5; sys_platform == 'linux'
|
24 |
+
WeTextProcessing; sys_platform == 'linux'
|
25 |
+
nemo_text_processing; sys_platform == 'linux'
|
26 |
+
|
27 |
vocos
|
28 |
|
29 |
spaces
|
tool/normalizer/__init__.py
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
from .normali_util import normalizer_en_nemo_text,normalizer_cn_tn
|
tool/normalizer/normali_util.py
ADDED
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Callable
|
2 |
+
from functools import partial
|
3 |
+
|
4 |
+
|
5 |
+
def normalizer_en_nemo_text() -> Callable[[str], str]:
|
6 |
+
from nemo_text_processing.text_normalization.normalize import Normalizer
|
7 |
+
|
8 |
+
return partial(
|
9 |
+
Normalizer(input_case="cased", lang="en").normalize,
|
10 |
+
verbose=False,
|
11 |
+
punct_post_process=True,
|
12 |
+
)
|
13 |
+
|
14 |
+
def normalizer_cn_tn() -> Callable[[str], str]:
|
15 |
+
from tn.chinese.normalizer import Normalizer
|
16 |
+
|
17 |
+
return Normalizer().normalize
|
web/app_gpu.py
CHANGED
@@ -96,6 +96,13 @@ def main(args):
|
|
96 |
interactive=True,
|
97 |
)
|
98 |
with gr.Row():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
voice_selection = gr.Dropdown(
|
100 |
label="Timbre",
|
101 |
choices=voices.keys(),
|
@@ -110,7 +117,6 @@ def main(args):
|
|
110 |
minimum=seed_min,
|
111 |
maximum=seed_max,
|
112 |
)
|
113 |
-
generate_audio_seed = gr.Button("随机生成音色种子", interactive=True)
|
114 |
text_seed_input = gr.Number(
|
115 |
value=42,
|
116 |
label="文本种子",
|
@@ -118,7 +124,9 @@ def main(args):
|
|
118 |
minimum=seed_min,
|
119 |
maximum=seed_max,
|
120 |
)
|
121 |
-
|
|
|
|
|
122 |
|
123 |
# with gr.Row():
|
124 |
# spk_emb_text = gr.Textbox(
|
@@ -172,7 +180,8 @@ def main(args):
|
|
172 |
temperature_slider,
|
173 |
top_p_slider,
|
174 |
top_k_slider,
|
175 |
-
audio_seed_input
|
|
|
176 |
],
|
177 |
outputs=[text_output,audio_output])
|
178 |
# 初始化 spk_emb_text 数值
|
@@ -206,7 +215,8 @@ def general_chat_infer_audio(text,
|
|
206 |
temperature_slider,
|
207 |
top_p_slider,
|
208 |
top_k_slider,
|
209 |
-
audio_seed_input
|
|
|
210 |
|
211 |
logger.info("========开始处理TTS模型=====")
|
212 |
#音频参数设置
|
@@ -229,6 +239,7 @@ def general_chat_infer_audio(text,
|
|
229 |
skip_refine_text=False,
|
230 |
refine_text_only=True, #仅返回优化后文本内容
|
231 |
params_refine_text=params_refine_text,
|
|
|
232 |
)
|
233 |
|
234 |
|
|
|
96 |
interactive=True,
|
97 |
)
|
98 |
with gr.Row():
|
99 |
+
lang_selection = gr.Dropdown(
|
100 |
+
label="语种",
|
101 |
+
choices=["zh" , "en"],
|
102 |
+
value="zh",
|
103 |
+
interactive=True,
|
104 |
+
show_label=True
|
105 |
+
)
|
106 |
voice_selection = gr.Dropdown(
|
107 |
label="Timbre",
|
108 |
choices=voices.keys(),
|
|
|
117 |
minimum=seed_min,
|
118 |
maximum=seed_max,
|
119 |
)
|
|
|
120 |
text_seed_input = gr.Number(
|
121 |
value=42,
|
122 |
label="文本种子",
|
|
|
124 |
minimum=seed_min,
|
125 |
maximum=seed_max,
|
126 |
)
|
127 |
+
with gr.Column():
|
128 |
+
generate_audio_seed = gr.Button("随机生成音色种子", interactive=True)
|
129 |
+
generate_text_seed = gr.Button("随机生成文本种子", interactive=True)
|
130 |
|
131 |
# with gr.Row():
|
132 |
# spk_emb_text = gr.Textbox(
|
|
|
180 |
temperature_slider,
|
181 |
top_p_slider,
|
182 |
top_k_slider,
|
183 |
+
audio_seed_input,
|
184 |
+
lang_selection
|
185 |
],
|
186 |
outputs=[text_output,audio_output])
|
187 |
# 初始化 spk_emb_text 数值
|
|
|
215 |
temperature_slider,
|
216 |
top_p_slider,
|
217 |
top_k_slider,
|
218 |
+
audio_seed_input,
|
219 |
+
lang):
|
220 |
|
221 |
logger.info("========开始处理TTS模型=====")
|
222 |
#音频参数设置
|
|
|
239 |
skip_refine_text=False,
|
240 |
refine_text_only=True, #仅返回优化后文本内容
|
241 |
params_refine_text=params_refine_text,
|
242 |
+
lang=lang
|
243 |
)
|
244 |
|
245 |
|