Spaces:
Runtime error
Runtime error
jianuo
commited on
Commit
•
c02528e
1
Parent(s):
454c9e4
支持输出报错的函数调用栈,并用更优雅的方式切换本地推理时的工作目录
Browse files- TTSs/base_tts.py +5 -1
- TTSs/genshin_local/genshin_bg.py +523 -497
- TTSs/genshin_local/genshin_local_tts.py +27 -20
TTSs/base_tts.py
CHANGED
@@ -3,6 +3,8 @@ import io
|
|
3 |
import scipy.io.wavfile as wavfile
|
4 |
from pydub import AudioSegment
|
5 |
|
|
|
|
|
6 |
import gradio as gr
|
7 |
|
8 |
|
@@ -136,6 +138,8 @@ class Base_TTS(metaclass=abc.ABCMeta):
|
|
136 |
|
137 |
return None, *mix_background_music(original_audio, 背景音乐, TTS_up, bg_up)
|
138 |
except Exception as e:
|
139 |
-
|
|
|
|
|
140 |
|
141 |
|
|
|
3 |
import scipy.io.wavfile as wavfile
|
4 |
from pydub import AudioSegment
|
5 |
|
6 |
+
import traceback
|
7 |
+
|
8 |
import gradio as gr
|
9 |
|
10 |
|
|
|
138 |
|
139 |
return None, *mix_background_music(original_audio, 背景音乐, TTS_up, bg_up)
|
140 |
except Exception as e:
|
141 |
+
msg = traceback.format_exc()
|
142 |
+
|
143 |
+
return msg + '\n\n' + str(e), None, None
|
144 |
|
145 |
|
TTSs/genshin_local/genshin_bg.py
CHANGED
@@ -1,161 +1,168 @@
|
|
1 |
# flake8: noqa: E402
|
2 |
import os
|
3 |
import sys
|
|
|
4 |
|
5 |
-
genshin_path = os.path.join(os.path.abspath(os.path.dirname(__file__)), 'genshin')
|
6 |
-
sys.path.append(os.path.abspath(os.path.dirname(__file__)))
|
7 |
-
sys.path.append(genshin_path)
|
8 |
-
os.chdir(genshin_path)
|
9 |
-
|
10 |
-
import genshin.re_matching as re_matching
|
11 |
-
from genshin.tools.sentence import split_by_language
|
12 |
-
|
13 |
-
import torch
|
14 |
-
import genshin.utils as utils
|
15 |
-
from genshin.infer import infer, latest_version, get_net_g, infer_multilang
|
16 |
import gradio as gr
|
17 |
-
import numpy as np
|
18 |
-
from genshin.config import config
|
19 |
-
from genshin.tools.translate import translate
|
20 |
import librosa
|
|
|
|
|
21 |
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
os.
|
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 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
159 |
speaker,
|
160 |
sdp_ratio,
|
161 |
noise_scale,
|
@@ -167,83 +174,119 @@ def tts_split(
|
|
167 |
style_text,
|
168 |
style_weight,
|
169 |
)
|
170 |
-
silence = np.zeros((int)(44100 *
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
def
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
for sentence, lang in sentences_list:
|
213 |
-
if sentence == "":
|
214 |
continue
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
228 |
-
|
229 |
-
|
230 |
-
|
231 |
-
emotion,
|
232 |
-
style_text=None,
|
233 |
-
style_weight=0,
|
234 |
-
):
|
235 |
-
audio_list = []
|
236 |
-
if language == "mix":
|
237 |
-
bool_valid, str_valid = re_matching.validate_text(text)
|
238 |
-
if not bool_valid:
|
239 |
-
return str_valid, (
|
240 |
-
hps.data.sampling_rate,
|
241 |
-
np.concatenate([np.zeros(hps.data.sampling_rate // 2)]),
|
242 |
-
)
|
243 |
-
for slice in re_matching.text_matching(text):
|
244 |
-
_text, _lang, _speaker = process_mix(slice)
|
245 |
-
if _speaker is None:
|
246 |
continue
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
247 |
print(f"Text: {_text}\nLang: {_lang}")
|
248 |
audio_list.extend(
|
249 |
generate_audio_multilang(
|
@@ -252,293 +295,276 @@ def process_text(
|
|
252 |
noise_scale,
|
253 |
noise_scale_w,
|
254 |
length_scale,
|
255 |
-
|
256 |
_lang,
|
257 |
reference_audio,
|
258 |
emotion,
|
259 |
)
|
260 |
)
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
)
|
277 |
-
else:
|
278 |
-
audio_list.extend(
|
279 |
-
generate_audio(
|
280 |
-
text.split("|"),
|
281 |
-
sdp_ratio,
|
282 |
-
noise_scale,
|
283 |
-
noise_scale_w,
|
284 |
-
length_scale,
|
285 |
-
speaker,
|
286 |
-
language,
|
287 |
-
reference_audio,
|
288 |
-
emotion,
|
289 |
-
style_text,
|
290 |
-
style_weight,
|
291 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
292 |
)
|
293 |
-
|
294 |
-
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
|
309 |
-
)
|
310 |
-
|
311 |
-
|
312 |
-
|
313 |
-
|
314 |
-
|
315 |
-
|
|
|
|
|
|
|
|
|
316 |
else:
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
|
321 |
-
audio_list = process_text(
|
322 |
-
text,
|
323 |
-
speaker,
|
324 |
-
sdp_ratio,
|
325 |
-
noise_scale,
|
326 |
-
noise_scale_w,
|
327 |
-
length_scale,
|
328 |
-
language,
|
329 |
-
reference_audio,
|
330 |
-
emotion,
|
331 |
-
style_text,
|
332 |
-
style_weight,
|
333 |
-
)
|
334 |
|
335 |
-
|
336 |
-
|
337 |
-
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
-
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
-
|
346 |
-
|
347 |
-
|
348 |
-
|
349 |
-
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
"visible": True,
|
364 |
-
"__type__": "update",
|
365 |
-
}
|
366 |
-
|
367 |
-
|
368 |
-
hps = utils.get_hparams_from_file(config.webui_config.config_path)
|
369 |
-
# 若config.json中未指定版本则默认为最新版本
|
370 |
-
version = hps.version if hasattr(hps, "version") else latest_version
|
371 |
-
net_g = get_net_g(
|
372 |
-
model_path=config.webui_config.model, version=version, device=device, hps=hps
|
373 |
-
)
|
374 |
-
speaker_ids = hps.data.spk2id
|
375 |
-
speakers = list(speaker_ids.keys())
|
376 |
-
languages = ["ZH", "JP", "EN", "mix", "auto"]
|
377 |
-
|
378 |
-
|
379 |
-
def get_advanced_block():
|
380 |
-
with gr.Blocks() as genshin_local:
|
381 |
-
gr.Markdown('原作者:https://www.bilibili.com/read/cv26659988/')
|
382 |
-
with gr.Row():
|
383 |
-
with gr.Column():
|
384 |
-
text = gr.TextArea(
|
385 |
-
label="输入文本内容",
|
386 |
-
placeholder="""
|
387 |
-
如果你选择语言为\'mix\',必须按照格式输入,否则报错:
|
388 |
-
格式举例(zh是中文,jp是日语,不区分大小写;说话人举例:gongzi):
|
389 |
-
[说话人1]<zh>你好,こんにちは! <jp>こんにちは,世界。
|
390 |
-
[说话人2]<zh>你好吗?<jp>元気ですか?
|
391 |
-
[说话人3]<zh>谢谢。<jp>どういたしまして。
|
392 |
-
...
|
393 |
-
另外,所有的语言选项都可以用'|'分割长段实现分句生成。
|
394 |
-
""",
|
395 |
-
)
|
396 |
-
trans = gr.Button("中翻日", variant="primary")
|
397 |
-
slicer = gr.Button("快速切分", variant="primary")
|
398 |
-
formatter = gr.Button("检测语言,并整理为 MIX 格式", variant="primary")
|
399 |
-
speaker = gr.Dropdown(
|
400 |
-
choices=speakers, value=speakers[0], label="Speaker"
|
401 |
-
)
|
402 |
-
_ = gr.Markdown(
|
403 |
-
value="提示模式(Prompt mode):可选文字提示或音频提示,用于生成文字或音频指定风格的声音。\n",
|
404 |
-
visible=False,
|
405 |
-
)
|
406 |
-
prompt_mode = gr.Radio(
|
407 |
-
["Text prompt", "Audio prompt"],
|
408 |
-
label="Prompt Mode",
|
409 |
-
value="Text prompt",
|
410 |
-
visible=False,
|
411 |
-
)
|
412 |
-
text_prompt = gr.Textbox(
|
413 |
-
label="Text prompt",
|
414 |
-
placeholder="用文字描述生成风格。如:Happy",
|
415 |
-
value="Happy",
|
416 |
-
visible=False,
|
417 |
-
)
|
418 |
-
audio_prompt = gr.Audio(
|
419 |
-
label="Audio prompt", type="filepath", visible=False
|
420 |
-
)
|
421 |
-
sdp_ratio = gr.Slider(
|
422 |
-
minimum=0, maximum=1, value=0.5, step=0.1, label="SDP Ratio"
|
423 |
-
)
|
424 |
-
noise_scale = gr.Slider(
|
425 |
-
minimum=0.1, maximum=2, value=0.6, step=0.1, label="Noise"
|
426 |
-
)
|
427 |
-
noise_scale_w = gr.Slider(
|
428 |
-
minimum=0.1, maximum=2, value=0.9, step=0.1, label="Noise_W"
|
429 |
-
)
|
430 |
-
length_scale = gr.Slider(
|
431 |
-
minimum=0.1, maximum=2, value=1.0, step=0.1, label="Length"
|
432 |
-
)
|
433 |
-
language = gr.Dropdown(
|
434 |
-
choices=languages, value=languages[0], label="Language"
|
435 |
-
)
|
436 |
-
btn = gr.Button("生成音频!", variant="primary")
|
437 |
-
with gr.Column():
|
438 |
-
with gr.Accordion("融合文本语义", open=False):
|
439 |
-
gr.Markdown(
|
440 |
-
value="使用辅助文本的语意来辅助生成对话(语言保持与主文本相同)\n\n"
|
441 |
-
"**注意**:不要使用**指令式文本**(如:开心),要使用**带有强烈情感的文本**(如:我好快乐!!!)\n\n"
|
442 |
-
"效果较不明确,留空即为不使用该功能"
|
443 |
)
|
444 |
-
|
445 |
-
|
446 |
-
|
447 |
-
|
448 |
-
value=0
|
449 |
-
step=0.1,
|
450 |
-
label="Weight",
|
451 |
-
info="主文本和辅助文本的bert混合比率,0表示仅主文本,1表示仅辅助文本",
|
452 |
)
|
453 |
-
|
454 |
-
|
455 |
-
|
456 |
-
|
457 |
-
|
458 |
-
|
459 |
-
|
460 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
461 |
)
|
462 |
-
|
|
|
463 |
minimum=0,
|
464 |
-
maximum=
|
465 |
-
value=
|
466 |
step=0.1,
|
467 |
-
label="
|
468 |
-
|
469 |
-
opt_cut_by_sent = gr.Checkbox(
|
470 |
-
label="按句切分 在按段落切分的基础上再按句子切分文本"
|
471 |
)
|
472 |
-
|
473 |
-
|
474 |
-
|
475 |
-
|
476 |
-
|
477 |
-
|
478 |
-
|
479 |
-
|
480 |
-
|
481 |
-
|
482 |
-
|
483 |
-
|
484 |
-
|
485 |
-
|
486 |
-
|
487 |
-
|
488 |
-
|
489 |
-
|
490 |
-
|
491 |
-
|
492 |
-
|
493 |
-
|
494 |
-
|
495 |
-
|
496 |
-
|
497 |
-
|
498 |
-
|
499 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
500 |
|
501 |
-
|
502 |
-
|
503 |
-
|
504 |
-
|
505 |
-
|
506 |
-
|
507 |
-
|
508 |
-
|
509 |
-
|
510 |
-
|
511 |
-
|
512 |
-
|
513 |
-
|
514 |
-
|
515 |
-
|
516 |
-
|
517 |
-
|
518 |
-
|
519 |
-
|
520 |
-
|
521 |
-
|
522 |
-
|
523 |
-
|
524 |
-
|
525 |
-
|
526 |
|
527 |
-
|
528 |
-
|
529 |
-
|
530 |
-
|
531 |
-
|
532 |
|
533 |
-
|
534 |
-
|
535 |
-
|
536 |
-
|
537 |
-
|
538 |
|
539 |
-
|
540 |
-
|
541 |
-
|
542 |
-
|
543 |
-
|
544 |
-
|
|
|
1 |
# flake8: noqa: E402
|
2 |
import os
|
3 |
import sys
|
4 |
+
from contextlib import contextmanager
|
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
import gradio as gr
|
|
|
|
|
|
|
7 |
import librosa
|
8 |
+
import numpy as np
|
9 |
+
import torch
|
10 |
|
11 |
+
|
12 |
+
@contextmanager
|
13 |
+
def change_dir():
|
14 |
+
file_path = os.path.abspath(os.path.dirname(__file__))
|
15 |
+
genshin_path = os.path.join(file_path, 'genshin')
|
16 |
+
|
17 |
+
need_rm = []
|
18 |
+
if file_path in sys.path:
|
19 |
+
need_rm.append(file_path)
|
20 |
+
if genshin_path in sys.path:
|
21 |
+
need_rm.append(genshin_path)
|
22 |
+
|
23 |
+
# 保存当前工作目录
|
24 |
+
current_dir = os.getcwd()
|
25 |
+
try:
|
26 |
+
os.chdir(genshin_path)
|
27 |
+
sys.path.append(file_path)
|
28 |
+
sys.path.append(genshin_path)
|
29 |
+
yield
|
30 |
+
finally:
|
31 |
+
os.chdir(current_dir)
|
32 |
+
for path in need_rm:
|
33 |
+
sys.path.remove(path)
|
34 |
+
|
35 |
+
|
36 |
+
# genshin_path = os.path.join(os.path.abspath(os.path.dirname(__file__)), 'genshin')
|
37 |
+
# sys.path.append(os.path.abspath(os.path.dirname(__file__)))
|
38 |
+
# sys.path.append(genshin_path)
|
39 |
+
# os.chdir(genshin_path)
|
40 |
+
|
41 |
+
with change_dir():
|
42 |
+
import genshin.re_matching as re_matching
|
43 |
+
from genshin.tools.sentence import split_by_language
|
44 |
+
import genshin.utils as utils
|
45 |
+
from genshin.infer import infer, latest_version, get_net_g, infer_multilang
|
46 |
+
from genshin.config import config
|
47 |
+
from genshin.tools.translate import translate
|
48 |
+
|
49 |
+
net_g = None
|
50 |
+
|
51 |
+
device = config.webui_config.device
|
52 |
+
if device == "mps":
|
53 |
+
os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
|
54 |
+
|
55 |
+
|
56 |
+
def generate_audio(
|
57 |
+
slices,
|
58 |
+
sdp_ratio,
|
59 |
+
noise_scale,
|
60 |
+
noise_scale_w,
|
61 |
+
length_scale,
|
62 |
+
speaker,
|
63 |
+
language,
|
64 |
+
reference_audio,
|
65 |
+
emotion,
|
66 |
+
style_text,
|
67 |
+
style_weight,
|
68 |
+
skip_start=False,
|
69 |
+
skip_end=False,
|
70 |
+
):
|
71 |
+
audio_list = []
|
72 |
+
# silence = np.zeros(hps.data.sampling_rate // 2, dtype=np.int16)
|
73 |
+
with torch.no_grad():
|
74 |
+
for idx, piece in enumerate(slices):
|
75 |
+
skip_start = idx != 0
|
76 |
+
skip_end = idx != len(slices) - 1
|
77 |
+
audio = infer(
|
78 |
+
piece,
|
79 |
+
reference_audio=reference_audio,
|
80 |
+
emotion=emotion,
|
81 |
+
sdp_ratio=sdp_ratio,
|
82 |
+
noise_scale=noise_scale,
|
83 |
+
noise_scale_w=noise_scale_w,
|
84 |
+
length_scale=length_scale,
|
85 |
+
sid=speaker,
|
86 |
+
language=language,
|
87 |
+
hps=hps,
|
88 |
+
net_g=net_g,
|
89 |
+
device=device,
|
90 |
+
skip_start=skip_start,
|
91 |
+
skip_end=skip_end,
|
92 |
+
style_text=style_text,
|
93 |
+
style_weight=style_weight,
|
94 |
+
)
|
95 |
+
audio16bit = gr.processing_utils.convert_to_16_bit_wav(audio)
|
96 |
+
audio_list.append(audio16bit)
|
97 |
+
return audio_list
|
98 |
+
|
99 |
+
|
100 |
+
def generate_audio_multilang(
|
101 |
+
slices,
|
102 |
+
sdp_ratio,
|
103 |
+
noise_scale,
|
104 |
+
noise_scale_w,
|
105 |
+
length_scale,
|
106 |
+
speaker,
|
107 |
+
language,
|
108 |
+
reference_audio,
|
109 |
+
emotion,
|
110 |
+
skip_start=False,
|
111 |
+
skip_end=False,
|
112 |
+
):
|
113 |
+
audio_list = []
|
114 |
+
# silence = np.zeros(hps.data.sampling_rate // 2, dtype=np.int16)
|
115 |
+
with torch.no_grad():
|
116 |
+
for idx, piece in enumerate(slices):
|
117 |
+
skip_start = idx != 0
|
118 |
+
skip_end = idx != len(slices) - 1
|
119 |
+
audio = infer_multilang(
|
120 |
+
piece,
|
121 |
+
reference_audio=reference_audio,
|
122 |
+
emotion=emotion,
|
123 |
+
sdp_ratio=sdp_ratio,
|
124 |
+
noise_scale=noise_scale,
|
125 |
+
noise_scale_w=noise_scale_w,
|
126 |
+
length_scale=length_scale,
|
127 |
+
sid=speaker,
|
128 |
+
language=language[idx],
|
129 |
+
hps=hps,
|
130 |
+
net_g=net_g,
|
131 |
+
device=device,
|
132 |
+
skip_start=skip_start,
|
133 |
+
skip_end=skip_end,
|
134 |
+
)
|
135 |
+
audio16bit = gr.processing_utils.convert_to_16_bit_wav(audio)
|
136 |
+
audio_list.append(audio16bit)
|
137 |
+
return audio_list
|
138 |
+
|
139 |
+
|
140 |
+
def tts_split(
|
141 |
+
text: str,
|
142 |
+
speaker,
|
143 |
+
sdp_ratio,
|
144 |
+
noise_scale,
|
145 |
+
noise_scale_w,
|
146 |
+
length_scale,
|
147 |
+
language,
|
148 |
+
cut_by_sent,
|
149 |
+
interval_between_para,
|
150 |
+
interval_between_sent,
|
151 |
+
reference_audio,
|
152 |
+
emotion,
|
153 |
+
style_text,
|
154 |
+
style_weight,
|
155 |
+
):
|
156 |
+
while text.find("\n\n") != -1:
|
157 |
+
text = text.replace("\n\n", "\n")
|
158 |
+
text = text.replace("|", "")
|
159 |
+
para_list = re_matching.cut_para(text)
|
160 |
+
para_list = [p for p in para_list if p != ""]
|
161 |
+
audio_list = []
|
162 |
+
for p in para_list:
|
163 |
+
if not cut_by_sent:
|
164 |
+
audio_list += process_text(
|
165 |
+
p,
|
166 |
speaker,
|
167 |
sdp_ratio,
|
168 |
noise_scale,
|
|
|
174 |
style_text,
|
175 |
style_weight,
|
176 |
)
|
177 |
+
silence = np.zeros((int)(44100 * interval_between_para), dtype=np.int16)
|
178 |
+
audio_list.append(silence)
|
179 |
+
else:
|
180 |
+
audio_list_sent = []
|
181 |
+
sent_list = re_matching.cut_sent(p)
|
182 |
+
sent_list = [s for s in sent_list if s != ""]
|
183 |
+
for s in sent_list:
|
184 |
+
audio_list_sent += process_text(
|
185 |
+
s,
|
186 |
+
speaker,
|
187 |
+
sdp_ratio,
|
188 |
+
noise_scale,
|
189 |
+
noise_scale_w,
|
190 |
+
length_scale,
|
191 |
+
language,
|
192 |
+
reference_audio,
|
193 |
+
emotion,
|
194 |
+
style_text,
|
195 |
+
style_weight,
|
196 |
+
)
|
197 |
+
silence = np.zeros((int)(44100 * interval_between_sent))
|
198 |
+
audio_list_sent.append(silence)
|
199 |
+
if (interval_between_para - interval_between_sent) > 0:
|
200 |
+
silence = np.zeros(
|
201 |
+
(int)(44100 * (interval_between_para - interval_between_sent))
|
202 |
+
)
|
203 |
+
audio_list_sent.append(silence)
|
204 |
+
audio16bit = gr.processing_utils.convert_to_16_bit_wav(
|
205 |
+
np.concatenate(audio_list_sent)
|
206 |
+
) # 对完整句子做音量归一
|
207 |
+
audio_list.append(audio16bit)
|
208 |
+
audio_concat = np.concatenate(audio_list)
|
209 |
+
return ("Success", (hps.data.sampling_rate, audio_concat))
|
210 |
+
|
211 |
+
|
212 |
+
def process_mix(slice):
|
213 |
+
_speaker = slice.pop()
|
214 |
+
_text, _lang = [], []
|
215 |
+
for lang, content in slice:
|
216 |
+
content = content.split("|")
|
217 |
+
content = [part for part in content if part != ""]
|
218 |
+
if len(content) == 0:
|
|
|
|
|
219 |
continue
|
220 |
+
if len(_text) == 0:
|
221 |
+
_text = [[part] for part in content]
|
222 |
+
_lang = [[lang] for part in content]
|
223 |
+
else:
|
224 |
+
_text[-1].append(content[0])
|
225 |
+
_lang[-1].append(lang)
|
226 |
+
if len(content) > 1:
|
227 |
+
_text += [[part] for part in content[1:]]
|
228 |
+
_lang += [[lang] for part in content[1:]]
|
229 |
+
return _text, _lang, _speaker
|
230 |
+
|
231 |
+
|
232 |
+
def process_auto(text):
|
233 |
+
_text, _lang = [], []
|
234 |
+
for slice in text.split("|"):
|
235 |
+
if slice == "":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
236 |
continue
|
237 |
+
temp_text, temp_lang = [], []
|
238 |
+
sentences_list = split_by_language(slice, target_languages=["zh", "ja", "en"])
|
239 |
+
for sentence, lang in sentences_list:
|
240 |
+
if sentence == "":
|
241 |
+
continue
|
242 |
+
temp_text.append(sentence)
|
243 |
+
temp_lang.append(lang.upper())
|
244 |
+
_text.append(temp_text)
|
245 |
+
_lang.append(temp_lang)
|
246 |
+
return _text, _lang
|
247 |
+
|
248 |
+
|
249 |
+
def process_text(
|
250 |
+
text: str,
|
251 |
+
speaker,
|
252 |
+
sdp_ratio,
|
253 |
+
noise_scale,
|
254 |
+
noise_scale_w,
|
255 |
+
length_scale,
|
256 |
+
language,
|
257 |
+
reference_audio,
|
258 |
+
emotion,
|
259 |
+
style_text=None,
|
260 |
+
style_weight=0,
|
261 |
+
):
|
262 |
+
audio_list = []
|
263 |
+
if language == "mix":
|
264 |
+
bool_valid, str_valid = re_matching.validate_text(text)
|
265 |
+
if not bool_valid:
|
266 |
+
return str_valid, (
|
267 |
+
hps.data.sampling_rate,
|
268 |
+
np.concatenate([np.zeros(hps.data.sampling_rate // 2)]),
|
269 |
+
)
|
270 |
+
for slice in re_matching.text_matching(text):
|
271 |
+
_text, _lang, _speaker = process_mix(slice)
|
272 |
+
if _speaker is None:
|
273 |
+
continue
|
274 |
+
print(f"Text: {_text}\nLang: {_lang}")
|
275 |
+
audio_list.extend(
|
276 |
+
generate_audio_multilang(
|
277 |
+
_text,
|
278 |
+
sdp_ratio,
|
279 |
+
noise_scale,
|
280 |
+
noise_scale_w,
|
281 |
+
length_scale,
|
282 |
+
_speaker,
|
283 |
+
_lang,
|
284 |
+
reference_audio,
|
285 |
+
emotion,
|
286 |
+
)
|
287 |
+
)
|
288 |
+
elif language.lower() == "auto":
|
289 |
+
_text, _lang = process_auto(text)
|
290 |
print(f"Text: {_text}\nLang: {_lang}")
|
291 |
audio_list.extend(
|
292 |
generate_audio_multilang(
|
|
|
295 |
noise_scale,
|
296 |
noise_scale_w,
|
297 |
length_scale,
|
298 |
+
speaker,
|
299 |
_lang,
|
300 |
reference_audio,
|
301 |
emotion,
|
302 |
)
|
303 |
)
|
304 |
+
else:
|
305 |
+
audio_list.extend(
|
306 |
+
generate_audio(
|
307 |
+
text.split("|"),
|
308 |
+
sdp_ratio,
|
309 |
+
noise_scale,
|
310 |
+
noise_scale_w,
|
311 |
+
length_scale,
|
312 |
+
speaker,
|
313 |
+
language,
|
314 |
+
reference_audio,
|
315 |
+
emotion,
|
316 |
+
style_text,
|
317 |
+
style_weight,
|
318 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
319 |
)
|
320 |
+
return audio_list
|
321 |
+
|
322 |
+
|
323 |
+
def tts_fn(
|
324 |
+
text: str,
|
325 |
+
speaker,
|
326 |
+
sdp_ratio,
|
327 |
+
noise_scale,
|
328 |
+
noise_scale_w,
|
329 |
+
length_scale,
|
330 |
+
language,
|
331 |
+
reference_audio,
|
332 |
+
emotion,
|
333 |
+
prompt_mode,
|
334 |
+
style_text=None,
|
335 |
+
style_weight=0,
|
336 |
+
):
|
337 |
+
if style_text == "":
|
338 |
+
style_text = None
|
339 |
+
if prompt_mode == "Audio prompt":
|
340 |
+
if reference_audio == None:
|
341 |
+
return ("Invalid audio prompt", None)
|
342 |
+
else:
|
343 |
+
reference_audio = load_audio(reference_audio)[1]
|
344 |
+
else:
|
345 |
+
reference_audio = None
|
346 |
+
|
347 |
+
audio_list = process_text(
|
348 |
+
text,
|
349 |
+
speaker,
|
350 |
+
sdp_ratio,
|
351 |
+
noise_scale,
|
352 |
+
noise_scale_w,
|
353 |
+
length_scale,
|
354 |
+
language,
|
355 |
+
reference_audio,
|
356 |
+
emotion,
|
357 |
+
style_text,
|
358 |
+
style_weight,
|
359 |
)
|
360 |
+
|
361 |
+
audio_concat = np.concatenate(audio_list)
|
362 |
+
return "Success", (hps.data.sampling_rate, audio_concat)
|
363 |
+
|
364 |
+
|
365 |
+
def format_utils(text, speaker):
|
366 |
+
_text, _lang = process_auto(text)
|
367 |
+
res = f"[{speaker}]"
|
368 |
+
for lang_s, content_s in zip(_lang, _text):
|
369 |
+
for lang, content in zip(lang_s, content_s):
|
370 |
+
res += f"<{lang.lower()}>{content}"
|
371 |
+
res += "|"
|
372 |
+
return "mix", res[:-1]
|
373 |
+
|
374 |
+
|
375 |
+
def load_audio(path):
|
376 |
+
audio, sr = librosa.load(path, 48000)
|
377 |
+
# audio = librosa.resample(audio, 44100, 48000)
|
378 |
+
return sr, audio
|
379 |
+
|
380 |
+
|
381 |
+
def gr_util(item):
|
382 |
+
if item == "Text prompt":
|
383 |
+
return {"visible": True, "__type__": "update"}, {
|
384 |
+
"visible": False,
|
385 |
+
"__type__": "update",
|
386 |
+
}
|
387 |
else:
|
388 |
+
return {"visible": False, "__type__": "update"}, {
|
389 |
+
"visible": True,
|
390 |
+
"__type__": "update",
|
391 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
392 |
|
393 |
+
|
394 |
+
hps = utils.get_hparams_from_file(config.webui_config.config_path)
|
395 |
+
# 若config.json中未指定版本则默认为最新版本
|
396 |
+
version = hps.version if hasattr(hps, "version") else latest_version
|
397 |
+
net_g = get_net_g(
|
398 |
+
model_path=config.webui_config.model, version=version, device=device, hps=hps
|
399 |
+
)
|
400 |
+
speaker_ids = hps.data.spk2id
|
401 |
+
speakers = list(speaker_ids.keys())
|
402 |
+
languages = ["ZH", "JP", "EN", "mix", "auto"]
|
403 |
+
|
404 |
+
|
405 |
+
def get_advanced_block():
|
406 |
+
with gr.Blocks() as genshin_local:
|
407 |
+
gr.Markdown('原作者:https://www.bilibili.com/read/cv26659988/')
|
408 |
+
with gr.Row():
|
409 |
+
with gr.Column():
|
410 |
+
text = gr.TextArea(
|
411 |
+
label="输入文本内容",
|
412 |
+
placeholder="""
|
413 |
+
如果你选择语言为\'mix\',必须按照格式输入,否则报错:
|
414 |
+
格式举例(zh是中文,jp是日语,不区分大小写;说话人举例:gongzi):
|
415 |
+
[说话人1]<zh>你好,こんにちは! <jp>こんにちは,世界。
|
416 |
+
[说话人2]<zh>你好吗?<jp>元気ですか?
|
417 |
+
[说话人3]<zh>谢谢。<jp>どういたしまして。
|
418 |
+
...
|
419 |
+
另外,所有的语言选项都可以用'|'分割长段实现分句生成。
|
420 |
+
""",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
421 |
)
|
422 |
+
trans = gr.Button("中翻日", variant="primary")
|
423 |
+
slicer = gr.Button("快速切分", variant="primary")
|
424 |
+
formatter = gr.Button("检测语言,并整理为 MIX 格式", variant="primary")
|
425 |
+
speaker = gr.Dropdown(
|
426 |
+
choices=speakers, value=speakers[0], label="Speaker"
|
|
|
|
|
|
|
427 |
)
|
428 |
+
_ = gr.Markdown(
|
429 |
+
value="提示模式(Prompt mode):可选文字提示或音频提示,用于生成文字或音频指定风格的声音。\n",
|
430 |
+
visible=False,
|
431 |
+
)
|
432 |
+
prompt_mode = gr.Radio(
|
433 |
+
["Text prompt", "Audio prompt"],
|
434 |
+
label="Prompt Mode",
|
435 |
+
value="Text prompt",
|
436 |
+
visible=False,
|
437 |
+
)
|
438 |
+
text_prompt = gr.Textbox(
|
439 |
+
label="Text prompt",
|
440 |
+
placeholder="用文字描述生成风格。如:Happy",
|
441 |
+
value="Happy",
|
442 |
+
visible=False,
|
443 |
+
)
|
444 |
+
audio_prompt = gr.Audio(
|
445 |
+
label="Audio prompt", type="filepath", visible=False
|
446 |
+
)
|
447 |
+
sdp_ratio = gr.Slider(
|
448 |
+
minimum=0, maximum=1, value=0.5, step=0.1, label="SDP Ratio"
|
449 |
+
)
|
450 |
+
noise_scale = gr.Slider(
|
451 |
+
minimum=0.1, maximum=2, value=0.6, step=0.1, label="Noise"
|
452 |
+
)
|
453 |
+
noise_scale_w = gr.Slider(
|
454 |
+
minimum=0.1, maximum=2, value=0.9, step=0.1, label="Noise_W"
|
455 |
+
)
|
456 |
+
length_scale = gr.Slider(
|
457 |
+
minimum=0.1, maximum=2, value=1.0, step=0.1, label="Length"
|
458 |
+
)
|
459 |
+
language = gr.Dropdown(
|
460 |
+
choices=languages, value=languages[0], label="Language"
|
461 |
+
)
|
462 |
+
btn = gr.Button("生成音频!", variant="primary")
|
463 |
+
with gr.Column():
|
464 |
+
with gr.Accordion("融合文本语义", open=False):
|
465 |
+
gr.Markdown(
|
466 |
+
value="使用辅助文本的语意来辅助生成对话(语言保持与主文本相同)\n\n"
|
467 |
+
"**注意**:不要使用**指令式文本**(如:开心),要使用**带有强烈情感的文本**(如:我好快乐!!!)\n\n"
|
468 |
+
"效果较不明确,留空即为不使用该功能"
|
469 |
)
|
470 |
+
style_text = gr.Textbox(label="辅助文本")
|
471 |
+
style_weight = gr.Slider(
|
472 |
minimum=0,
|
473 |
+
maximum=1,
|
474 |
+
value=0.7,
|
475 |
step=0.1,
|
476 |
+
label="Weight",
|
477 |
+
info="主文本和辅助文本的bert混合比率,0表示仅主文本,1表示仅辅助文本",
|
|
|
|
|
478 |
)
|
479 |
+
with gr.Row():
|
480 |
+
with gr.Column():
|
481 |
+
interval_between_sent = gr.Slider(
|
482 |
+
minimum=0,
|
483 |
+
maximum=5,
|
484 |
+
value=0.2,
|
485 |
+
step=0.1,
|
486 |
+
label="句间停顿(秒),勾选按句切分才生效",
|
487 |
+
)
|
488 |
+
interval_between_para = gr.Slider(
|
489 |
+
minimum=0,
|
490 |
+
maximum=10,
|
491 |
+
value=1,
|
492 |
+
step=0.1,
|
493 |
+
label="段间停顿(秒),需要大于句间停顿才有效",
|
494 |
+
)
|
495 |
+
opt_cut_by_sent = gr.Checkbox(
|
496 |
+
label="按句切分 在按段落切分的基础上再按句子切分文本"
|
497 |
+
)
|
498 |
+
slicer = gr.Button("切分生成", variant="primary")
|
499 |
+
text_output = gr.Textbox(label="状态信息")
|
500 |
+
audio_output = gr.Audio(label="输出音频")
|
501 |
+
# explain_image = gr.Image(
|
502 |
+
# label="参数解释信息",
|
503 |
+
# show_label=True,
|
504 |
+
# show_share_button=False,
|
505 |
+
# show_download_button=False,
|
506 |
+
# value=os.path.abspath("./img/参数说明.png"),
|
507 |
+
# )
|
508 |
+
btn.click(
|
509 |
+
tts_fn,
|
510 |
+
inputs=[
|
511 |
+
text,
|
512 |
+
speaker,
|
513 |
+
sdp_ratio,
|
514 |
+
noise_scale,
|
515 |
+
noise_scale_w,
|
516 |
+
length_scale,
|
517 |
+
language,
|
518 |
+
audio_prompt,
|
519 |
+
text_prompt,
|
520 |
+
prompt_mode,
|
521 |
+
style_text,
|
522 |
+
style_weight,
|
523 |
+
],
|
524 |
+
outputs=[text_output, audio_output],
|
525 |
+
)
|
526 |
|
527 |
+
trans.click(
|
528 |
+
translate,
|
529 |
+
inputs=[text],
|
530 |
+
outputs=[text],
|
531 |
+
)
|
532 |
+
slicer.click(
|
533 |
+
tts_split,
|
534 |
+
inputs=[
|
535 |
+
text,
|
536 |
+
speaker,
|
537 |
+
sdp_ratio,
|
538 |
+
noise_scale,
|
539 |
+
noise_scale_w,
|
540 |
+
length_scale,
|
541 |
+
language,
|
542 |
+
opt_cut_by_sent,
|
543 |
+
interval_between_para,
|
544 |
+
interval_between_sent,
|
545 |
+
audio_prompt,
|
546 |
+
text_prompt,
|
547 |
+
style_text,
|
548 |
+
style_weight,
|
549 |
+
],
|
550 |
+
outputs=[text_output, audio_output],
|
551 |
+
)
|
552 |
|
553 |
+
prompt_mode.change(
|
554 |
+
lambda x: gr_util(x),
|
555 |
+
inputs=[prompt_mode],
|
556 |
+
outputs=[text_prompt, audio_prompt],
|
557 |
+
)
|
558 |
|
559 |
+
audio_prompt.upload(
|
560 |
+
lambda x: load_audio(x),
|
561 |
+
inputs=[audio_prompt],
|
562 |
+
outputs=[audio_prompt],
|
563 |
+
)
|
564 |
|
565 |
+
formatter.click(
|
566 |
+
format_utils,
|
567 |
+
inputs=[text, speaker],
|
568 |
+
outputs=[language, text],
|
569 |
+
)
|
570 |
+
return genshin_local
|
TTSs/genshin_local/genshin_local_tts.py
CHANGED
@@ -1,27 +1,33 @@
|
|
1 |
import io
|
|
|
2 |
import os
|
|
|
3 |
|
4 |
import gradio as gr
|
5 |
import scipy.io.wavfile as wavfile
|
6 |
from pydub import AudioSegment
|
7 |
|
8 |
-
import importlib
|
9 |
-
import logging
|
10 |
-
|
11 |
from TTSs.base_tts import Base_TTS
|
12 |
|
13 |
|
14 |
-
|
15 |
class genshin_local_TTS(Base_TTS):
|
16 |
-
|
17 |
def __init__(self):
|
18 |
if self.is_show():
|
19 |
try:
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
logging.info('导入原神本地语音合成模块成功')
|
24 |
except Exception as e:
|
|
|
|
|
25 |
logging.error('导入原神本地语音合成模块失败')
|
26 |
logging.error(e)
|
27 |
else:
|
@@ -95,16 +101,18 @@ class genshin_local_TTS(Base_TTS):
|
|
95 |
noise_scale_w_local,
|
96 |
length_scale_local,
|
97 |
language_local):
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
|
|
|
|
108 |
|
109 |
wav_io = io.BytesIO()
|
110 |
wavfile.write(wav_io, ori_audio_data[0], ori_audio_data[1])
|
@@ -112,11 +120,10 @@ class genshin_local_TTS(Base_TTS):
|
|
112 |
original_audio = AudioSegment.from_wav(wav_io)
|
113 |
return original_audio
|
114 |
|
115 |
-
|
116 |
def search_speaker(self, search_value):
|
117 |
for s in self.speakers_genshin_local:
|
118 |
if search_value == s:
|
119 |
return s
|
120 |
for s in self.speakers_genshin_local:
|
121 |
if search_value in s:
|
122 |
-
return s
|
|
|
1 |
import io
|
2 |
+
import logging
|
3 |
import os
|
4 |
+
import traceback
|
5 |
|
6 |
import gradio as gr
|
7 |
import scipy.io.wavfile as wavfile
|
8 |
from pydub import AudioSegment
|
9 |
|
|
|
|
|
|
|
10 |
from TTSs.base_tts import Base_TTS
|
11 |
|
12 |
|
|
|
13 |
class genshin_local_TTS(Base_TTS):
|
|
|
14 |
def __init__(self):
|
15 |
if self.is_show():
|
16 |
try:
|
17 |
+
from .genshin_bg import speakers
|
18 |
+
from .genshin_bg import languages
|
19 |
+
from .genshin_bg import tts_fn
|
20 |
+
from .genshin_bg import change_dir
|
21 |
+
|
22 |
+
self.speakers_genshin_local = speakers
|
23 |
+
self.languages = languages
|
24 |
+
self.genshin_tts_fn = tts_fn
|
25 |
+
self.change_dir = change_dir
|
26 |
+
|
27 |
logging.info('导入原神本地语音合成模块成功')
|
28 |
except Exception as e:
|
29 |
+
traceback.print_exc()
|
30 |
+
|
31 |
logging.error('导入原神本地语音合成模块失败')
|
32 |
logging.error(e)
|
33 |
else:
|
|
|
101 |
noise_scale_w_local,
|
102 |
length_scale_local,
|
103 |
language_local):
|
104 |
+
|
105 |
+
with self.change_dir():
|
106 |
+
ori_audio_data = self.genshin_tts_fn(text, speaker_local,
|
107 |
+
sdp_ratio_local,
|
108 |
+
noise_scale_local,
|
109 |
+
noise_scale_w_local,
|
110 |
+
length_scale_local,
|
111 |
+
language_local,
|
112 |
+
None, 'Happy',
|
113 |
+
'Text prompt',
|
114 |
+
'style_text',
|
115 |
+
0.7)[1]
|
116 |
|
117 |
wav_io = io.BytesIO()
|
118 |
wavfile.write(wav_io, ori_audio_data[0], ori_audio_data[1])
|
|
|
120 |
original_audio = AudioSegment.from_wav(wav_io)
|
121 |
return original_audio
|
122 |
|
|
|
123 |
def search_speaker(self, search_value):
|
124 |
for s in self.speakers_genshin_local:
|
125 |
if search_value == s:
|
126 |
return s
|
127 |
for s in self.speakers_genshin_local:
|
128 |
if search_value in s:
|
129 |
+
return s
|