raoyonghui
commited on
Commit
•
a8db66d
1
Parent(s):
0491f05
auto detect prompt language and text
Browse files
app.py
CHANGED
@@ -19,23 +19,44 @@ from models.tts.maskgct.g2p.g2p_generation import g2p, chn_eng_g2p
|
|
19 |
|
20 |
from transformers import SeamlessM4TFeatureExtractor
|
21 |
|
22 |
-
|
23 |
|
24 |
processor = SeamlessM4TFeatureExtractor.from_pretrained("facebook/w2v-bert-2.0")
|
25 |
|
26 |
device = torch.device("cuda" if torch.cuda.is_available() else "CPU")
|
|
|
27 |
|
28 |
-
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
#
|
31 |
-
|
32 |
-
|
33 |
-
#
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
|
40 |
|
41 |
def g2p_(text, language):
|
@@ -279,9 +300,7 @@ def load_models():
|
|
279 |
@torch.no_grad()
|
280 |
def maskgct_inference(
|
281 |
prompt_speech_path,
|
282 |
-
prompt_text,
|
283 |
target_text,
|
284 |
-
language="en",
|
285 |
target_language="en",
|
286 |
target_len=None,
|
287 |
n_timesteps=25,
|
@@ -295,14 +314,17 @@ def maskgct_inference(
|
|
295 |
speech_16k = librosa.load(prompt_speech_path, sr=16000)[0]
|
296 |
speech = librosa.load(prompt_speech_path, sr=24000)[0]
|
297 |
|
298 |
-
|
299 |
-
|
300 |
-
|
|
|
|
|
|
|
301 |
combine_semantic_code, _ = text2semantic(
|
302 |
device,
|
303 |
speech_16k,
|
304 |
-
|
305 |
-
|
306 |
target_text,
|
307 |
target_language,
|
308 |
target_len,
|
@@ -326,20 +348,16 @@ def maskgct_inference(
|
|
326 |
@spaces.GPU
|
327 |
def inference(
|
328 |
prompt_wav,
|
329 |
-
prompt_text,
|
330 |
target_text,
|
331 |
target_len,
|
332 |
n_timesteps,
|
333 |
-
language,
|
334 |
target_language,
|
335 |
):
|
336 |
save_path = "./output/output.wav"
|
337 |
os.makedirs("./output", exist_ok=True)
|
338 |
recovered_audio = maskgct_inference(
|
339 |
prompt_wav,
|
340 |
-
prompt_text,
|
341 |
target_text,
|
342 |
-
language,
|
343 |
target_language,
|
344 |
target_len=target_len,
|
345 |
n_timesteps=int(n_timesteps),
|
@@ -369,7 +387,6 @@ iface = gr.Interface(
|
|
369 |
fn=inference,
|
370 |
inputs=[
|
371 |
gr.Audio(label="Upload Prompt Wav", type="filepath"),
|
372 |
-
gr.Textbox(label="Prompt Text"),
|
373 |
gr.Textbox(label="Target Text"),
|
374 |
gr.Number(
|
375 |
label="Target Duration (in seconds), if the target duration is less than 0, the system will estimate a duration.", value=-1
|
@@ -377,7 +394,6 @@ iface = gr.Interface(
|
|
377 |
gr.Slider(
|
378 |
label="Number of Timesteps", minimum=15, maximum=100, value=25, step=1
|
379 |
),
|
380 |
-
gr.Dropdown(label="Language", choices=language_list, value="en"),
|
381 |
gr.Dropdown(label="Target Language", choices=language_list, value="en"),
|
382 |
],
|
383 |
outputs=gr.Audio(label="Generated Audio"),
|
|
|
19 |
|
20 |
from transformers import SeamlessM4TFeatureExtractor
|
21 |
|
22 |
+
import whisper
|
23 |
|
24 |
processor = SeamlessM4TFeatureExtractor.from_pretrained("facebook/w2v-bert-2.0")
|
25 |
|
26 |
device = torch.device("cuda" if torch.cuda.is_available() else "CPU")
|
27 |
+
whisper_model = whisper.load_model("turbo")
|
28 |
|
29 |
+
def detect_speech_language(speech_file):
|
30 |
+
# load audio and pad/trim it to fit 30 seconds
|
31 |
+
whisper_model = whisper.load_model("turbo")
|
32 |
+
audio = whisper.load_audio(speech_file)
|
33 |
+
audio = whisper.pad_or_trim(audio)
|
34 |
|
35 |
+
# make log-Mel spectrogram and move to the same device as the model
|
36 |
+
mel = whisper.log_mel_spectrogram(audio, n_mels=128).to(whisper_model.device)
|
37 |
+
|
38 |
+
# detect the spoken language
|
39 |
+
_, probs = whisper_model.detect_language(mel)
|
40 |
+
return max(probs, key=probs.get)
|
41 |
+
|
42 |
+
|
43 |
+
@torch.no_grad()
|
44 |
+
def get_prompt_text(speech_16k, language):
|
45 |
+
full_prompt_text = ""
|
46 |
+
shot_prompt_text = ""
|
47 |
+
short_prompt_end_ts = 0.0
|
48 |
+
|
49 |
+
asr_result = whisper_model.transcribe(speech_16k, language=language)
|
50 |
+
full_prompt_text = asr_result["text"] # whisper asr result
|
51 |
+
#text = asr_result["segments"][0]["text"] # whisperx asr result
|
52 |
+
shot_prompt_text = ""
|
53 |
+
short_prompt_end_ts = 0.0
|
54 |
+
for segment in asr_result["segments"]:
|
55 |
+
shot_prompt_text = shot_prompt_text + segment['text']
|
56 |
+
short_prompt_end_ts = segment['end']
|
57 |
+
if short_prompt_end_ts >= 4:
|
58 |
+
break
|
59 |
+
return full_prompt_text, shot_prompt_text, short_prompt_end_ts
|
60 |
|
61 |
|
62 |
def g2p_(text, language):
|
|
|
300 |
@torch.no_grad()
|
301 |
def maskgct_inference(
|
302 |
prompt_speech_path,
|
|
|
303 |
target_text,
|
|
|
304 |
target_language="en",
|
305 |
target_len=None,
|
306 |
n_timesteps=25,
|
|
|
314 |
speech_16k = librosa.load(prompt_speech_path, sr=16000)[0]
|
315 |
speech = librosa.load(prompt_speech_path, sr=24000)[0]
|
316 |
|
317 |
+
prompt_language = detect_speech_language(prompt_speech_path)
|
318 |
+
full_prompt_text, short_prompt_text, shot_prompt_end_ts = get_prompt_text(prompt_speech_path,
|
319 |
+
prompt_language)
|
320 |
+
# use the first 4+ seconds wav as the prompt in case the prompt wav is too long
|
321 |
+
speech = speech[0: int(shot_prompt_end_ts * 24000)]
|
322 |
+
speech_16k = speech_16k[0: int(shot_prompt_end_ts*16000)]
|
323 |
combine_semantic_code, _ = text2semantic(
|
324 |
device,
|
325 |
speech_16k,
|
326 |
+
short_prompt_text,
|
327 |
+
prompt_language,
|
328 |
target_text,
|
329 |
target_language,
|
330 |
target_len,
|
|
|
348 |
@spaces.GPU
|
349 |
def inference(
|
350 |
prompt_wav,
|
|
|
351 |
target_text,
|
352 |
target_len,
|
353 |
n_timesteps,
|
|
|
354 |
target_language,
|
355 |
):
|
356 |
save_path = "./output/output.wav"
|
357 |
os.makedirs("./output", exist_ok=True)
|
358 |
recovered_audio = maskgct_inference(
|
359 |
prompt_wav,
|
|
|
360 |
target_text,
|
|
|
361 |
target_language,
|
362 |
target_len=target_len,
|
363 |
n_timesteps=int(n_timesteps),
|
|
|
387 |
fn=inference,
|
388 |
inputs=[
|
389 |
gr.Audio(label="Upload Prompt Wav", type="filepath"),
|
|
|
390 |
gr.Textbox(label="Target Text"),
|
391 |
gr.Number(
|
392 |
label="Target Duration (in seconds), if the target duration is less than 0, the system will estimate a duration.", value=-1
|
|
|
394 |
gr.Slider(
|
395 |
label="Number of Timesteps", minimum=15, maximum=100, value=25, step=1
|
396 |
),
|
|
|
397 |
gr.Dropdown(label="Target Language", choices=language_list, value="en"),
|
398 |
],
|
399 |
outputs=gr.Audio(label="Generated Audio"),
|