Replaced Encodec with Vocos
Browse files- app.py +56 -62
- requirements.txt +1 -0
app.py
CHANGED
@@ -1,4 +1,3 @@
|
|
1 |
-
import argparse
|
2 |
import logging
|
3 |
import os
|
4 |
import pathlib
|
@@ -19,7 +18,6 @@ langid.set_languages(['en', 'zh', 'ja'])
|
|
19 |
|
20 |
import torch
|
21 |
import torchaudio
|
22 |
-
import random
|
23 |
|
24 |
import numpy as np
|
25 |
|
@@ -35,7 +33,8 @@ from macros import *
|
|
35 |
from examples import *
|
36 |
|
37 |
import gradio as gr
|
38 |
-
import
|
|
|
39 |
|
40 |
torch._C._jit_set_profiling_executor(False)
|
41 |
torch._C._jit_set_profiling_mode(False)
|
@@ -72,8 +71,13 @@ model.eval()
|
|
72 |
# Encodec model
|
73 |
audio_tokenizer = AudioTokenizer(device)
|
74 |
|
|
|
|
|
|
|
75 |
# ASR
|
76 |
-
|
|
|
|
|
77 |
|
78 |
# Voice Presets
|
79 |
preset_list = os.walk("./presets/").__next__()[2]
|
@@ -89,34 +93,33 @@ def clear_prompts():
|
|
89 |
endfiletime = time.time() - 60
|
90 |
if endfiletime > lastmodifytime:
|
91 |
os.remove(filename)
|
|
|
|
|
92 |
except:
|
93 |
return
|
94 |
|
95 |
-
def transcribe_one(
|
96 |
-
|
97 |
-
|
98 |
-
|
|
|
99 |
|
100 |
-
|
101 |
-
mel = whisper.log_mel_spectrogram(audio).to(model.device)
|
102 |
|
103 |
-
#
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
options = whisper.DecodingOptions(temperature=1.0, best_of=5, fp16=False if device == torch.device("cpu") else True, sample_len=150)
|
109 |
-
result = whisper.decode(model, mel, options)
|
110 |
|
111 |
# print the recognized text
|
112 |
-
print(
|
113 |
|
114 |
-
text_pr = result.text
|
115 |
if text_pr.strip(" ")[-1] not in "?!.,。,?!。、":
|
116 |
text_pr += "."
|
117 |
|
118 |
# delete all variables
|
119 |
-
del
|
120 |
gc.collect()
|
121 |
return lang, text_pr
|
122 |
|
@@ -137,7 +140,7 @@ def make_npz_prompt(name, uploaded_audio, recorded_audio, transcript_content):
|
|
137 |
assert wav_pr.ndim and wav_pr.size(0) == 1
|
138 |
|
139 |
if transcript_content == "":
|
140 |
-
|
141 |
else:
|
142 |
lang_pr = langid.classify(str(transcript_content))[0]
|
143 |
lang_token = lang2token[lang_pr]
|
@@ -147,6 +150,8 @@ def make_npz_prompt(name, uploaded_audio, recorded_audio, transcript_content):
|
|
147 |
audio_tokens = encoded_frames[0][0].transpose(2, 1).cpu().numpy()
|
148 |
|
149 |
# tokenize text
|
|
|
|
|
150 |
phonemes, _ = text_tokenizer.tokenize(text=f"{text_pr}".strip())
|
151 |
text_tokens, enroll_x_lens = text_collater(
|
152 |
[
|
@@ -155,6 +160,8 @@ def make_npz_prompt(name, uploaded_audio, recorded_audio, transcript_content):
|
|
155 |
)
|
156 |
|
157 |
message = f"Detected language: {lang_pr}\n Detected text {text_pr}\n"
|
|
|
|
|
158 |
|
159 |
# save as npz file
|
160 |
np.savez(os.path.join(tempfile.gettempdir(), f"{name}.npz"),
|
@@ -166,30 +173,6 @@ def make_npz_prompt(name, uploaded_audio, recorded_audio, transcript_content):
|
|
166 |
return message, os.path.join(tempfile.gettempdir(), f"{name}.npz")
|
167 |
|
168 |
|
169 |
-
def make_prompt(name, wav, sr, save=True):
|
170 |
-
if not isinstance(wav, torch.FloatTensor):
|
171 |
-
wav = torch.tensor(wav)
|
172 |
-
if wav.abs().max() > 1:
|
173 |
-
wav /= wav.abs().max()
|
174 |
-
if wav.size(-1) == 2:
|
175 |
-
wav = wav.mean(-1, keepdim=False)
|
176 |
-
if wav.ndim == 1:
|
177 |
-
wav = wav.unsqueeze(0)
|
178 |
-
assert wav.ndim and wav.size(0) == 1
|
179 |
-
torchaudio.save(f"./prompts/{name}.wav", wav, sr)
|
180 |
-
lang, text = transcribe_one(whisper_model, f"./prompts/{name}.wav")
|
181 |
-
lang_token = lang2token[lang]
|
182 |
-
text = lang_token + text + lang_token
|
183 |
-
with open(f"./prompts/{name}.txt", 'w') as f:
|
184 |
-
f.write(text)
|
185 |
-
if not save:
|
186 |
-
os.remove(f"./prompts/{name}.wav")
|
187 |
-
os.remove(f"./prompts/{name}.txt")
|
188 |
-
# delete all variables
|
189 |
-
del lang_token, wav, sr
|
190 |
-
gc.collect()
|
191 |
-
return text, lang
|
192 |
-
|
193 |
@torch.no_grad()
|
194 |
def infer_from_audio(text, language, accent, audio_prompt, record_audio_prompt, transcript_content):
|
195 |
if len(text) > 150:
|
@@ -209,7 +192,7 @@ def infer_from_audio(text, language, accent, audio_prompt, record_audio_prompt,
|
|
209 |
assert wav_pr.ndim and wav_pr.size(0) == 1
|
210 |
|
211 |
if transcript_content == "":
|
212 |
-
|
213 |
else:
|
214 |
lang_pr = langid.classify(str(transcript_content))[0]
|
215 |
lang_token = lang2token[lang_pr]
|
@@ -222,6 +205,9 @@ def infer_from_audio(text, language, accent, audio_prompt, record_audio_prompt,
|
|
222 |
lang = token2lang[lang_token]
|
223 |
text = lang_token + text + lang_token
|
224 |
|
|
|
|
|
|
|
225 |
# tokenize audio
|
226 |
encoded_frames = tokenize_audio(audio_tokenizer, (wav_pr, sr))
|
227 |
audio_prompts = encoded_frames[0][0].transpose(2, 1).to(device)
|
@@ -237,6 +223,8 @@ def infer_from_audio(text, language, accent, audio_prompt, record_audio_prompt,
|
|
237 |
|
238 |
enroll_x_lens = None
|
239 |
if text_pr:
|
|
|
|
|
240 |
text_prompts, _ = text_tokenizer.tokenize(text=f"{text_pr}".strip())
|
241 |
text_prompts, enroll_x_lens = text_collater(
|
242 |
[
|
@@ -256,15 +244,16 @@ def infer_from_audio(text, language, accent, audio_prompt, record_audio_prompt,
|
|
256 |
prompt_language=lang_pr,
|
257 |
text_language=langs if accent == "no-accent" else lang,
|
258 |
)
|
259 |
-
|
260 |
-
|
261 |
-
)
|
|
|
262 |
|
263 |
message = f"text prompt: {text_pr}\nsythesized text: {text}"
|
264 |
# delete all variables
|
265 |
del audio_prompts, text_tokens, text_prompts, phone_tokens, encoded_frames, wav_pr, sr, audio_prompt, record_audio_prompt, transcript_content
|
266 |
gc.collect()
|
267 |
-
return message, (24000, samples
|
268 |
|
269 |
@torch.no_grad()
|
270 |
def infer_from_prompt(text, language, accent, preset_prompt, prompt_file):
|
@@ -315,16 +304,17 @@ def infer_from_prompt(text, language, accent, preset_prompt, prompt_file):
|
|
315 |
prompt_language=lang_pr,
|
316 |
text_language=langs if accent == "no-accent" else lang,
|
317 |
)
|
318 |
-
|
319 |
-
|
320 |
-
)
|
|
|
321 |
|
322 |
message = f"sythesized text: {text}"
|
323 |
|
324 |
# delete all variables
|
325 |
del audio_prompts, text_tokens, text_prompts, phone_tokens, encoded_frames, prompt_file, preset_prompt
|
326 |
gc.collect()
|
327 |
-
return message, (24000, samples
|
328 |
|
329 |
|
330 |
from utils.sentence_cutter import split_text_into_sentences
|
@@ -407,11 +397,13 @@ def infer_long_text(text, preset_prompt, prompt=None, language='auto', accent='n
|
|
407 |
text_language=langs if accent == "no-accent" else lang,
|
408 |
)
|
409 |
complete_tokens = torch.cat([complete_tokens, encoded_frames.transpose(2, 1)], dim=-1)
|
410 |
-
|
411 |
-
|
412 |
-
)
|
|
|
|
|
413 |
message = f"Cut into {len(sentences)} sentences"
|
414 |
-
return message, (24000, samples
|
415 |
elif mode == "sliding-window":
|
416 |
complete_tokens = torch.zeros([1, NUM_QUANTIZERS, 0]).type(torch.LongTensor).to(device)
|
417 |
original_audio_prompts = audio_prompts
|
@@ -453,12 +445,14 @@ def infer_long_text(text, preset_prompt, prompt=None, language='auto', accent='n
|
|
453 |
else:
|
454 |
audio_prompts = original_audio_prompts
|
455 |
text_prompts = original_text_prompts
|
456 |
-
|
457 |
-
|
458 |
-
)
|
|
|
|
|
459 |
message = f"Cut into {len(sentences)} sentences"
|
460 |
|
461 |
-
return message, (24000, samples
|
462 |
else:
|
463 |
raise ValueError(f"No such mode {mode}")
|
464 |
|
|
|
|
|
1 |
import logging
|
2 |
import os
|
3 |
import pathlib
|
|
|
18 |
|
19 |
import torch
|
20 |
import torchaudio
|
|
|
21 |
|
22 |
import numpy as np
|
23 |
|
|
|
33 |
from examples import *
|
34 |
|
35 |
import gradio as gr
|
36 |
+
from vocos import Vocos
|
37 |
+
from transformers import WhisperProcessor, WhisperForConditionalGeneration
|
38 |
|
39 |
torch._C._jit_set_profiling_executor(False)
|
40 |
torch._C._jit_set_profiling_mode(False)
|
|
|
71 |
# Encodec model
|
72 |
audio_tokenizer = AudioTokenizer(device)
|
73 |
|
74 |
+
# Vocos decoder
|
75 |
+
vocos = Vocos.from_pretrained('charactr/vocos-encodec-24khz').to(device)
|
76 |
+
|
77 |
# ASR
|
78 |
+
whisper_processor = WhisperProcessor.from_pretrained("openai/whisper-medium")
|
79 |
+
whisper = WhisperForConditionalGeneration.from_pretrained("openai/whisper-medium").to(device)
|
80 |
+
whisper.config.forced_decoder_ids = None
|
81 |
|
82 |
# Voice Presets
|
83 |
preset_list = os.walk("./presets/").__next__()[2]
|
|
|
93 |
endfiletime = time.time() - 60
|
94 |
if endfiletime > lastmodifytime:
|
95 |
os.remove(filename)
|
96 |
+
del path, filename, lastmodifytime, endfiletime
|
97 |
+
gc.collect()
|
98 |
except:
|
99 |
return
|
100 |
|
101 |
+
def transcribe_one(wav, sr):
|
102 |
+
if sr != 16000:
|
103 |
+
wav4trans = torchaudio.transforms.Resample(sr, 16000)(wav)
|
104 |
+
else:
|
105 |
+
wav4trans = wav
|
106 |
|
107 |
+
input_features = whisper_processor(wav4trans.squeeze(0), sampling_rate=16000, return_tensors="pt").input_features
|
|
|
108 |
|
109 |
+
# generate token ids
|
110 |
+
predicted_ids = whisper.generate(input_features.to(device))
|
111 |
+
lang = whisper_processor.batch_decode(predicted_ids[:, 1])[0].strip("<|>")
|
112 |
+
# decode token ids to text
|
113 |
+
text_pr = whisper_processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
|
|
|
|
|
114 |
|
115 |
# print the recognized text
|
116 |
+
print(text_pr)
|
117 |
|
|
|
118 |
if text_pr.strip(" ")[-1] not in "?!.,。,?!。、":
|
119 |
text_pr += "."
|
120 |
|
121 |
# delete all variables
|
122 |
+
del wav4trans, input_features, predicted_ids
|
123 |
gc.collect()
|
124 |
return lang, text_pr
|
125 |
|
|
|
140 |
assert wav_pr.ndim and wav_pr.size(0) == 1
|
141 |
|
142 |
if transcript_content == "":
|
143 |
+
lang_pr, text_pr = transcribe_one(wav_pr, sr)
|
144 |
else:
|
145 |
lang_pr = langid.classify(str(transcript_content))[0]
|
146 |
lang_token = lang2token[lang_pr]
|
|
|
150 |
audio_tokens = encoded_frames[0][0].transpose(2, 1).cpu().numpy()
|
151 |
|
152 |
# tokenize text
|
153 |
+
lang_token = lang2token[lang_pr]
|
154 |
+
text_pr = lang_token + text_pr + lang_token
|
155 |
phonemes, _ = text_tokenizer.tokenize(text=f"{text_pr}".strip())
|
156 |
text_tokens, enroll_x_lens = text_collater(
|
157 |
[
|
|
|
160 |
)
|
161 |
|
162 |
message = f"Detected language: {lang_pr}\n Detected text {text_pr}\n"
|
163 |
+
if lang_pr not in ['ja', 'zh', 'en']:
|
164 |
+
return f"Prompt can only made with one of model-supported languages, got {lang_pr} instead", None
|
165 |
|
166 |
# save as npz file
|
167 |
np.savez(os.path.join(tempfile.gettempdir(), f"{name}.npz"),
|
|
|
173 |
return message, os.path.join(tempfile.gettempdir(), f"{name}.npz")
|
174 |
|
175 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
176 |
@torch.no_grad()
|
177 |
def infer_from_audio(text, language, accent, audio_prompt, record_audio_prompt, transcript_content):
|
178 |
if len(text) > 150:
|
|
|
192 |
assert wav_pr.ndim and wav_pr.size(0) == 1
|
193 |
|
194 |
if transcript_content == "":
|
195 |
+
lang_pr, text_pr = transcribe_one(wav_pr, sr)
|
196 |
else:
|
197 |
lang_pr = langid.classify(str(transcript_content))[0]
|
198 |
lang_token = lang2token[lang_pr]
|
|
|
205 |
lang = token2lang[lang_token]
|
206 |
text = lang_token + text + lang_token
|
207 |
|
208 |
+
if lang_pr not in ['ja', 'zh', 'en']:
|
209 |
+
return f"Reference audio must be a speech of one of model-supported languages, got {lang_pr} instead", None
|
210 |
+
|
211 |
# tokenize audio
|
212 |
encoded_frames = tokenize_audio(audio_tokenizer, (wav_pr, sr))
|
213 |
audio_prompts = encoded_frames[0][0].transpose(2, 1).to(device)
|
|
|
223 |
|
224 |
enroll_x_lens = None
|
225 |
if text_pr:
|
226 |
+
lang_token = lang2token[lang_pr]
|
227 |
+
text_pr = lang_token + text_pr + lang_token
|
228 |
text_prompts, _ = text_tokenizer.tokenize(text=f"{text_pr}".strip())
|
229 |
text_prompts, enroll_x_lens = text_collater(
|
230 |
[
|
|
|
244 |
prompt_language=lang_pr,
|
245 |
text_language=langs if accent == "no-accent" else lang,
|
246 |
)
|
247 |
+
# Decode with Vocos
|
248 |
+
frames = encoded_frames.permute(2,0,1)
|
249 |
+
features = vocos.codes_to_features(frames)
|
250 |
+
samples = vocos.decode(features, bandwidth_id=torch.tensor([2], device=device))
|
251 |
|
252 |
message = f"text prompt: {text_pr}\nsythesized text: {text}"
|
253 |
# delete all variables
|
254 |
del audio_prompts, text_tokens, text_prompts, phone_tokens, encoded_frames, wav_pr, sr, audio_prompt, record_audio_prompt, transcript_content
|
255 |
gc.collect()
|
256 |
+
return message, (24000, samples.squeeze(0).cpu().numpy())
|
257 |
|
258 |
@torch.no_grad()
|
259 |
def infer_from_prompt(text, language, accent, preset_prompt, prompt_file):
|
|
|
304 |
prompt_language=lang_pr,
|
305 |
text_language=langs if accent == "no-accent" else lang,
|
306 |
)
|
307 |
+
# Decode with Vocos
|
308 |
+
frames = encoded_frames.permute(2,0,1)
|
309 |
+
features = vocos.codes_to_features(frames)
|
310 |
+
samples = vocos.decode(features, bandwidth_id=torch.tensor([2], device=device))
|
311 |
|
312 |
message = f"sythesized text: {text}"
|
313 |
|
314 |
# delete all variables
|
315 |
del audio_prompts, text_tokens, text_prompts, phone_tokens, encoded_frames, prompt_file, preset_prompt
|
316 |
gc.collect()
|
317 |
+
return message, (24000, samples.squeeze(0).cpu().numpy())
|
318 |
|
319 |
|
320 |
from utils.sentence_cutter import split_text_into_sentences
|
|
|
397 |
text_language=langs if accent == "no-accent" else lang,
|
398 |
)
|
399 |
complete_tokens = torch.cat([complete_tokens, encoded_frames.transpose(2, 1)], dim=-1)
|
400 |
+
# Decode with Vocos
|
401 |
+
frames = encoded_frames.permute(2, 0, 1)
|
402 |
+
features = vocos.codes_to_features(frames)
|
403 |
+
samples = vocos.decode(features, bandwidth_id=torch.tensor([2], device=device))
|
404 |
+
|
405 |
message = f"Cut into {len(sentences)} sentences"
|
406 |
+
return message, (24000, samples.squeeze(0).cpu().numpy())
|
407 |
elif mode == "sliding-window":
|
408 |
complete_tokens = torch.zeros([1, NUM_QUANTIZERS, 0]).type(torch.LongTensor).to(device)
|
409 |
original_audio_prompts = audio_prompts
|
|
|
445 |
else:
|
446 |
audio_prompts = original_audio_prompts
|
447 |
text_prompts = original_text_prompts
|
448 |
+
# Decode with Vocos
|
449 |
+
frames = encoded_frames.permute(2, 0, 1)
|
450 |
+
features = vocos.codes_to_features(frames)
|
451 |
+
samples = vocos.decode(features, bandwidth_id=torch.tensor([2], device=device))
|
452 |
+
|
453 |
message = f"Cut into {len(sentences)} sentences"
|
454 |
|
455 |
+
return message, (24000, samples.squeeze(0).cpu().numpy())
|
456 |
else:
|
457 |
raise ValueError(f"No such mode {mode}")
|
458 |
|
requirements.txt
CHANGED
@@ -5,6 +5,7 @@ torchvision==0.15.2
|
|
5 |
torchaudio
|
6 |
tokenizers
|
7 |
encodec
|
|
|
8 |
langid
|
9 |
unidecode
|
10 |
pyopenjtalk
|
|
|
5 |
torchaudio
|
6 |
tokenizers
|
7 |
encodec
|
8 |
+
vocos
|
9 |
langid
|
10 |
unidecode
|
11 |
pyopenjtalk
|