-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathdub.py
1306 lines (1088 loc) · 49.1 KB
/
dub.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
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
160
161
162
163
164
165
166
167
168
169
170
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
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
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
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
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
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
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
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
base_path = "."
# base_path = "/content"
#@title import Model
# import locale
# locale.getpreferredencoding = lambda: "UTF-8"
import os
install_path=f"{base_path}/"
os.chdir(install_path)
import os
import re
import torch
import torchaudio
from einops import rearrange
from vocos import Vocos
from model import CFM, UNetT, DiT
from model.utils import (
load_checkpoint,
get_tokenizer,
convert_char_to_pinyin,
save_spectrogram,
)
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
import numpy as np
import librosa
import nltk
nltk.download('punkt')
from nltk.tokenize import sent_tokenize
from pydub import AudioSegment
import re
import uuid
from tqdm.notebook import tqdm
import shutil
from IPython.display import clear_output
import gc
import time
import subprocess
from IPython.display import Audio
import torch
def get_max_gpu_memory():
if torch.cuda.is_available():
gpu_id = torch.cuda.current_device()
max_memory = torch.cuda.get_device_properties(gpu_id).total_memory
max_memory_gb = max_memory / (1024 ** 3) # Convert bytes to GB
return max_memory_gb -1
else:
return 0 # No GPU available
def is_gpu_memory_over_limit(limit_gb=14.5):
limit_gb=get_max_gpu_memory()
# Run nvidia-smi and capture the output
result = subprocess.run(['nvidia-smi', '--query-gpu=memory.used', '--format=csv,nounits,noheader'],
stdout=subprocess.PIPE, text=True)
# Split the result into lines (for each GPU if there are multiple)
memory_used_mb_list = result.stdout.strip().splitlines()
# Convert memory used from MB to GB and check each GPU's memory usage
for i, memory_used_mb in enumerate(memory_used_mb_list):
memory_used_gb = int(memory_used_mb) / 1024.0
# print(f"GPU {i}: Current memory allocated: {memory_used_gb:.2f} GB")
if memory_used_gb > limit_gb:
# print(f"GPU {i} memory usage exceeds {limit_gb} GB.")
return True
# print("GPU memory usage is within safe limits.")
return False
# Load Whisper model
def load_whisper():
global whisper_pipe,whisper_model
try:
if whisper_pipe is not None:
del whisper_pipe
whisper_pipe=None
if whisper_model is not None:
del whisper_model
whisper_model=None
gc.collect()
torch.cuda.empty_cache()
# print("Free GPU memeory")
time.sleep(2)
except:
pass
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
model_id = "openai/whisper-large-v3-turbo"
whisper_model = AutoModelForSpeechSeq2Seq.from_pretrained(
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
)
whisper_model.to(device)
processor = AutoProcessor.from_pretrained(model_id)
whisper_pipe = pipeline(
"automatic-speech-recognition",
model=whisper_model,
tokenizer=processor.tokenizer,
feature_extractor=processor.feature_extractor,
torch_dtype=torch_dtype,
device=device,
)
return whisper_pipe,whisper_model
# Initialize vocoder and model function
def initialize_vocoder_and_model(
exp_name="F5TTS_Base",
ckpt_step=1200000,
device="cuda",
target_sample_rate=24000,
n_mel_channels=100,
hop_length=256,
dataset_name="Emilia_ZH_EN",
tokenizer="pinyin",
ode_method='euler',
use_ema=True
):
global vocos,model
try:
if vocos is not None:
del vocos
vocos=None
if model is not None:
del model
model=None
gc.collect()
torch.cuda.empty_cache()
# print("Free GPU memeory")
time.sleep(2)
except:
pass
# Set model configuration based on experiment name
if exp_name == "F5TTS_Base":
model_cls = DiT
model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)
elif exp_name == "E2TTS_Base":
model_cls = UNetT
model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)
# Initialize vocoder
vocos_local_path = "./ckpts/vocos-mel-24khz"
vocos = Vocos.from_hparams(f"{vocos_local_path}/config.yaml")
state_dict = torch.load(f"{vocos_local_path}/pytorch_model.bin", map_location=device)
vocos.load_state_dict(state_dict)
vocos.eval()
# Initialize tokenizer
vocab_char_map, vocab_size = get_tokenizer(dataset_name, tokenizer)
# Initialize model
model = CFM(
transformer=model_cls(
**model_cfg,
text_num_embeds=vocab_size,
mel_dim=n_mel_channels
),
mel_spec_kwargs=dict(
target_sample_rate=target_sample_rate,
n_mel_channels=n_mel_channels,
hop_length=hop_length,
),
odeint_kwargs=dict(
method=ode_method,
),
vocab_char_map=vocab_char_map,
).to(device)
# Load the model checkpoint
ckpt_path = f"ckpts/{exp_name}/model_{ckpt_step}.pt"
model = load_checkpoint(model, ckpt_path, device, use_ema=use_ema)
return vocos, model
def merge_audio(audio_list, save_path):
# Initialize an empty audio segment
merged_audio = AudioSegment.empty()
# Loop through the list of audio files
for audio_file in audio_list:
# Load each audio file
audio_segment = AudioSegment.from_wav(audio_file)
# Append to the merged audio segment
merged_audio += audio_segment
# Export the merged audio to the specified save path
merged_audio.export(save_path, format="wav")
def chunks_sentences(paragraph, join_limit=2):
sentences = sent_tokenize(paragraph)
# Initialize an empty list to store the new sentences
new_sentences = []
# Iterate through the list of sentences in steps of 'join_limit'
for i in range(0, len(sentences), join_limit):
# Join the sentences with a space between them
new_sentence = ' '.join(sentences[i:i + join_limit])
new_sentences.append(new_sentence)
return new_sentences
def clean_file_name(file_path):
# Get the base file name and extension
file_name = os.path.basename(file_path)
file_name, file_extension = os.path.splitext(file_name)
# Replace non-alphanumeric characters with an underscore
cleaned = re.sub(r'[^a-zA-Z\d]+', '_', file_name)
# Remove any multiple underscores
clean_file_name = re.sub(r'_+', '_', cleaned).strip('_')
# Generate a random UUID for uniqueness
random_uuid = uuid.uuid4().hex[:6]
# Combine cleaned file name with the original extension
clean_file_path = os.path.join(os.path.dirname(file_path), clean_file_name + f"_{random_uuid}" + file_extension)
return clean_file_path
def tts_file_name(text):
if text.endswith("."):
text = text[:-1]
text = text.lower()
text = text.strip()
text = text.replace(" ","_")
truncated_text = text[:25] if len(text) > 25 else text if len(text) > 0 else "empty"
random_string = uuid.uuid4().hex[:8].upper()
file_name = f"{base_path}/f5_Voice/{truncated_text}_{random_string}.wav"
file_name=clean_file_name(file_name)
return file_name
import os
from pydub import AudioSegment
def is_audio_duration_greater_than_30s(audio_path,max_duration=30):
try:
audio = AudioSegment.from_file(audio_path)
except Exception as e:
print(f"Error loading audio file: {e}")
return False
# Get the duration in seconds
duration = len(audio) / 1000 # pydub works in milliseconds
# Check if the duration is greater than 30 seconds
return duration > max_duration
def trim_audio(input_audio_path,max_duration=30):
# Create output folder if it doesn't exist
global base_path
output_folder=f"{base_path}/trim_audio"
os.makedirs(output_folder, exist_ok=True)
# Load the audio file
audio = AudioSegment.from_file(input_audio_path)
# Check the duration of the audio in seconds
duration = len(audio) / 1000 # pydub works in milliseconds
# Trim the audio if it exceeds the max_duration
if duration > max_duration:
trimmed_audio = audio[:max_duration * 1000] # Trim to max_duration
else:
trimmed_audio = audio
# Generate a new file name
base_name = os.path.splitext(os.path.basename(input_audio_path))[0]
output_file = f"{output_folder}/{base_name}_trimmed.wav"
# output_file=clean_file_name(output_file)
trimmed_audio.export(output_file, format="wav")
return output_file
def process_audio(reference_audio, max_duration=15):
global old_trim_audio,base_path
# Check if the audio duration exceeds max_duration
if is_audio_duration_greater_than_30s(reference_audio, max_duration):
f_base_name = os.path.basename(reference_audio)
f_name, f_extension = os.path.splitext(f_base_name)
trimmed_audio_path = f"{base_path}/trim_audio/{f_name}_trimmed.wav"
# Check if we've already trimmed this audio
if old_trim_audio == trimmed_audio_path:
reference_audio = trimmed_audio_path # Use existing trimmed audio
# print("skipping because same trim audio")
else:
reference_audio = trim_audio(reference_audio, max_duration) # Trim the audio
old_trim_audio = reference_audio # Update the old trimmed audio path
return reference_audio
# Voice cloning function
def voice_clone(reference_audio, text, output_dir="", target_sample_rate=24000,remove_silence = False,fix_duration=None,chunks=0,exp_name="F5TTS_Base",progress_bar=True):
global device,old_audio_path,old_ref_text,old_exp_name
global whisper_pipe,whisper_model,vocos,model
global seed
reference_audio=process_audio(reference_audio, max_duration=15)
if old_exp_name==exp_name:
pass
else:
vocos, model = initialize_vocoder_and_model(device=device,exp_name=exp_name)
old_exp_name=exp_name
if is_gpu_memory_over_limit():
whisper_pipe,whisper_model = load_whisper()
device = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
vocos, model = initialize_vocoder_and_model(device=device,exp_name=exp_name)
# clear_output()
# seed = None
final_audio_path= tts_file_name(text)
output_dir=f"{base_path}/f5_Voice/temp"
nfe_step = 32
cfg_strength = 2.
ode_method = 'euler' # euler | midpoint
speed = 1.
# target_sample_rate = 24000
# fix_duration=27
target_rms = 0.1
hop_length = 256 # Ensure hop_length is define
tokenizer="pinyin"
sway_sampling_coef = -1.
fix_duration = fix_duration # None (will linear estimate. if code-switched, consider fix) | float (total in seconds, include ref audio)
# Get the reference audio text
if old_audio_path==reference_audio:
ref_text=old_ref_text
# print("skipping because same audio file")
else:
ref_text = whisper_pipe(reference_audio)['text'].strip()
old_audio_path=reference_audio
old_ref_text=ref_text
# Ensure output directory exists
if os.path.exists(output_dir):
shutil.rmtree(output_dir)
os.makedirs(output_dir)
# Load audio
audio, sr = torchaudio.load(reference_audio)
if audio.shape[0] > 1:
audio = torch.mean(audio, dim=0, keepdim=True)
# Normalize audio
rms = torch.sqrt(torch.mean(torch.square(audio)))
if rms < target_rms:
audio = audio * target_rms / rms
# Resample audio if necessary
if sr != target_sample_rate:
resampler = torchaudio.transforms.Resample(sr, target_sample_rate)
audio = resampler(audio)
audio = audio.to(device)
if chunks==0:
prompts=[text]
else:
prompts=chunks_sentences(text, join_limit=chunks)
audio_list=[]
number_of_prompts=len(prompts)
iterable = tqdm(enumerate(prompts), total=len(prompts), desc="Processing Prompts") if progress_bar else enumerate(prompts)
# for i,text in enumerate(prompts):
# for i, text in tqdm(enumerate(prompts), total=len(prompts), desc="Processing Prompts"):
for i, text in iterable:
gen_text = text.strip()
# Prepare text
text_list = [ref_text + gen_text]
if tokenizer == "pinyin":
final_text_list = convert_char_to_pinyin(text_list)
else:
final_text_list = [text_list]
# print(f"text : {text_list}")
# print(f"pinyin: {final_text_list}")
# Calculate duration
ref_audio_len = audio.shape[-1] // hop_length
if fix_duration is not None:
if number_of_prompts==1:
duration = int(fix_duration * target_sample_rate / hop_length)
else:
zh_pause_punc = r"。，、；：？！"
ref_text_len = len(ref_text) + len(re.findall(zh_pause_punc, ref_text))
gen_text_len = len(gen_text) + len(re.findall(zh_pause_punc, gen_text))
duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)
# Inference
with torch.inference_mode():
generated, _ = model.sample(
cond=audio,
text=final_text_list,
duration=duration,
steps=nfe_step,
cfg_strength=cfg_strength,
sway_sampling_coef=sway_sampling_coef,
seed=seed,
)
generated = generated[:, ref_audio_len:, :]
generated_mel_spec = rearrange(generated, '1 n d -> 1 d n')
generated_wave = vocos.decode(generated_mel_spec.cpu())
if rms < target_rms:
generated_wave = generated_wave * rms / target_rms
if isinstance(generated_wave, torch.Tensor):
generated_wave = generated_wave.squeeze().cpu().numpy()
# Remove silence
if remove_silence:
# Detect non-silent intervals
non_silent_intervals = librosa.effects.split(generated_wave, top_db=30)
# Concatenate non-silent parts
non_silent_wave = np.array([])
for interval in non_silent_intervals:
start, end = interval
non_silent_wave = np.concatenate([non_silent_wave, generated_wave[start:end]])
# Replace generated_wave with the non-silent version
generated_wave = non_silent_wave
generated_wave_tensor = torch.tensor(generated_wave).unsqueeze(0)
else:
generated_wave_tensor = torch.tensor(generated_wave).unsqueeze(0)
# Ensure that generated_wave_tensor is 2D (batch size, channels)
if len(generated_wave_tensor.shape) == 1:
generated_wave_tensor = generated_wave_tensor.unsqueeze(0)
# Save the generated audio
save_audio_path = f"{output_dir}/{i}.wav"
torchaudio.save(save_audio_path, generated_wave_tensor, target_sample_rate)
audio_list.append(save_audio_path)
if len(audio_list)==1:
shutil.copy(audio_list[-1],final_audio_path)
elif len(audio_list)>1:
merge_audio(audio_list, final_audio_path)
else:
final_audio_path=None
return final_audio_path
whisper_pipe = None
whisper_model=None
vocos = None
model = None
seed = None
whisper_pipe,whisper_model = load_whisper()
device = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
vocos, model = initialize_vocoder_and_model(device=device,exp_name="F5TTS_Base")
old_audio_path=""
old_ref_text=""
old_trim_audio=""
old_exp_name="F5TTS_Base"
os.makedirs(f"{base_path}/f5_Voice", exist_ok=True)
clear_output()
print("Model Import Complete")
# #@title Generate TTS
# Reference_Audio_Path= '/content/F5-TTS/tests/ref_audio/test_en_1_ref_short.wav' # @param {type: "string"}
# TTS_Text = 'A cat named Luna loved watching the rain from her window. One day, a small bird landed on the sill, chirping happily despite the storm. Luna, curious but gentle, tapped the glass with her paw. The bird fluttered away, leaving Luna to wonder what it was like to fly. As the rain continued, she dreamed of soaring through the clouds.' # @param {type: "string"}
# Remove_Silence = True # @param {type: "boolean"}
# # Split_Sentences = 0 # @param {type: "number"}
# if len(TTS_Text)<=135:
# Split_Sentences=0
# if len(TTS_Text)>135:
# Split_Sentences=2
# Choose_Model = "F5TTS_Base" # @param ['F5TTS_Base', 'E2TTS_Base']
# seed = None
# cloned_voice_path=voice_clone(Reference_Audio_Path, TTS_Text,remove_silence = Remove_Silence,chunks=Split_Sentences,exp_name=Choose_Model)
# clear_output()
# print(f"TTS Save at {cloned_voice_path}")
# Audio(cloned_voice_path)
#@title Utils
language_dict = {
"Akan": {"lang_code": "aka", "meta_code": "aka_Latn"},
"Albanian": {"lang_code": "sq", "meta_code": "als_Latn"},
"Amharic": {"lang_code": "am", "meta_code": "amh_Ethi"},
"Arabic": {"lang_code": "ar", "meta_code": "arb_Arab"},
"Armenian": {"lang_code": "hy", "meta_code": "hye_Armn"},
"Assamese": {"lang_code": "as", "meta_code": "asm_Beng"},
"Azerbaijani": {"lang_code": "az", "meta_code": "azj_Latn"},
"Basque": {"lang_code": "eu", "meta_code": "eus_Latn"},
"Bashkir": {"lang_code": "ba", "meta_code": "bak_Cyrl"},
"Bengali": {"lang_code": "bn", "meta_code": "ben_Beng"},
"Bosnian": {"lang_code": "bs", "meta_code": "bos_Latn"},
"Bulgarian": {"lang_code": "bg", "meta_code": "bul_Cyrl"},
"Burmese": {"lang_code": "my", "meta_code": "mya_Mymr"},
"Catalan": {"lang_code": "ca", "meta_code": "cat_Latn"},
"Chinese": {"lang_code": "zh", "meta_code": "zh_Hans"},
"Croatian": {"lang_code": "hr", "meta_code": "hrv_Latn"},
"Czech": {"lang_code": "cs", "meta_code": "ces_Latn"},
"Danish": {"lang_code": "da", "meta_code": "dan_Latn"},
"Dutch": {"lang_code": "nl", "meta_code": "nld_Latn"},
"English": {"lang_code": "en", "meta_code": "eng_Latn"},
"Estonian": {"lang_code": "et", "meta_code": "est_Latn"},
"Faroese": {"lang_code": "fo", "meta_code": "fao_Latn"},
"Finnish": {"lang_code": "fi", "meta_code": "fin_Latn"},
"French": {"lang_code": "fr", "meta_code": "fra_Latn"},
"Galician": {"lang_code": "gl", "meta_code": "glg_Latn"},
"Georgian": {"lang_code": "ka", "meta_code": "kat_Geor"},
"German": {"lang_code": "de", "meta_code": "deu_Latn"},
"Greek": {"lang_code": "el", "meta_code": "ell_Grek"},
"Gujarati": {"lang_code": "gu", "meta_code": "guj_Gujr"},
"Haitian Creole": {"lang_code": "ht", "meta_code": "hat_Latn"},
"Hausa": {"lang_code": "ha", "meta_code": "hau_Latn"},
"Hebrew": {"lang_code": "he", "meta_code": "heb_Hebr"},
"Hindi": {"lang_code": "hi", "meta_code": "hin_Deva"},
"Hungarian": {"lang_code": "hu", "meta_code": "hun_Latn"},
"Icelandic": {"lang_code": "is", "meta_code": "isl_Latn"},
"Indonesian": {"lang_code": "id", "meta_code": "ind_Latn"},
"Italian": {"lang_code": "it", "meta_code": "ita_Latn"},
"Japanese": {"lang_code": "ja", "meta_code": "jpn_Jpan"},
"Kannada": {"lang_code": "kn", "meta_code": "kan_Knda"},
"Kazakh": {"lang_code": "kk", "meta_code": "kaz_Cyrl"},
"Korean": {"lang_code": "ko", "meta_code": "kor_Hang"},
"Kurdish": {"lang_code": "ckb", "meta_code": "ckb_Arab"},
"Kyrgyz": {"lang_code": "ky", "meta_code": "kir_Cyrl"},
"Lao": {"lang_code": "lo", "meta_code": "lao_Laoo"},
"Lithuanian": {"lang_code": "lt", "meta_code": "lit_Latn"},
"Luxembourgish": {"lang_code": "lb", "meta_code": "ltz_Latn"},
"Macedonian": {"lang_code": "mk", "meta_code": "mkd_Cyrl"},
"Malay": {"lang_code": "ms", "meta_code": "ms_Latn"},
"Malayalam": {"lang_code": "ml", "meta_code": "mal_Mlym"},
"Maltese": {"lang_code": "mt", "meta_code": "mlt_Latn"},
"Maori": {"lang_code": "mi", "meta_code": "mri_Latn"},
"Marathi": {"lang_code": "mr", "meta_code": "mar_Deva"},
"Mongolian": {"lang_code": "mn", "meta_code": "khk_Cyrl"},
"Nepali": {"lang_code": "ne", "meta_code": "npi_Deva"},
"Norwegian": {"lang_code": "no", "meta_code": "nob_Latn"},
"Norwegian Nynorsk": {"lang_code": "nn", "meta_code": "nno_Latn"},
"Pashto": {"lang_code": "ps", "meta_code": "pbt_Arab"},
"Persian": {"lang_code": "fa", "meta_code": "pes_Arab"},
"Polish": {"lang_code": "pl", "meta_code": "pol_Latn"},
"Portuguese": {"lang_code": "pt", "meta_code": "por_Latn"},
"Punjabi": {"lang_code": "pa", "meta_code": "pan_Guru"},
"Romanian": {"lang_code": "ro", "meta_code": "ron_Latn"},
"Russian": {"lang_code": "ru", "meta_code": "rus_Cyrl"},
"Serbian": {"lang_code": "sr", "meta_code": "srp_Cyrl"},
"Sinhala": {"lang_code": "si", "meta_code": "sin_Sinh"},
"Slovak": {"lang_code": "sk", "meta_code": "slk_Latn"},
"Slovenian": {"lang_code": "sl", "meta_code": "slv_Latn"},
"Somali": {"lang_code": "so", "meta_code": "som_Latn"},
"Spanish": {"lang_code": "es", "meta_code": "spa_Latn"},
"Sundanese": {"lang_code": "su", "meta_code": "sun_Latn"},
"Swahili": {"lang_code": "sw", "meta_code": "swa_Latn"},
"Swedish": {"lang_code": "sv", "meta_code": "swe_Latn"},
"Tamil": {"lang_code": "ta", "meta_code": "tam_Taml"},
"Telugu": {"lang_code": "te", "meta_code": "tel_Telu"},
"Thai": {"lang_code": "th", "meta_code": "tha_Latn"},
"Turkish": {"lang_code": "tr", "meta_code": "tur_Latn"},
"Ukrainian": {"lang_code": "uk", "meta_code": "ukr_Cyrl"},
"Urdu": {"lang_code": "ur", "meta_code": "urd_Arab"},
"Uzbek": {"lang_code": "uz", "meta_code": "uzb_Latn"},
"Vietnamese": {"lang_code": "vi", "meta_code": "vie_Latn"},
"Welsh": {"lang_code": "cy", "meta_code": "cym_Latn"},
"Yiddish": {"lang_code": "yi", "meta_code": "yi_Hebr"},
"Yoruba": {"lang_code": "yo", "meta_code": "yo_Latn"},
"Zulu": {"lang_code": "zu", "meta_code": "zul_Latn"},
}
available_language=['English','Hindi','Bengali','Akan', 'Albanian', 'Amharic', 'Arabic', 'Armenian', 'Assamese', 'Azerbaijani', 'Basque', 'Bashkir', 'Bengali', 'Bosnian', 'Bulgarian', 'Burmese', 'Catalan', 'Chinese', 'Croatian', 'Czech', 'Danish', 'Dutch', 'English', 'Estonian', 'Faroese', 'Finnish', 'French', 'Galician', 'Georgian', 'German', 'Greek', 'Gujarati', 'Haitian Creole', 'Hausa', 'Hebrew', 'Hindi', 'Hungarian', 'Icelandic', 'Indonesian', 'Italian', 'Japanese', 'Kannada', 'Kazakh', 'Korean', 'Kurdish', 'Kyrgyz', 'Lao', 'Lithuanian', 'Luxembourgish', 'Macedonian', 'Malay', 'Malayalam', 'Maltese', 'Maori', 'Marathi', 'Mongolian', 'Nepali', 'Norwegian', 'Norwegian Nynorsk', 'Pashto', 'Persian', 'Polish', 'Portuguese', 'Punjabi', 'Romanian', 'Russian', 'Serbian', 'Sinhala', 'Slovak', 'Slovenian', 'Somali', 'Spanish', 'Sundanese', 'Swahili', 'Swedish', 'Tamil', 'Telugu', 'Thai', 'Turkish', 'Ukrainian', 'Urdu', 'Uzbek', 'Vietnamese', 'Welsh', 'Yiddish', 'Yoruba', 'Zulu']
import math
import torch
import gc
import time
import subprocess
from faster_whisper import WhisperModel
import os
import mimetypes
import shutil
import re
import uuid
from pydub import AudioSegment
from transformers import pipeline
def get_language_name(lang_code):
global language_dict
# Iterate through the language dictionary
for language, details in language_dict.items():
# Check if the language code matches
if details["lang_code"] == lang_code:
return language # Return the language name
return None
def clean_file_name(file_path):
# Get the base file name and extension
file_name = os.path.basename(file_path)
file_name, file_extension = os.path.splitext(file_name)
# Replace non-alphanumeric characters with an underscore
cleaned = re.sub(r'[^a-zA-Z\d]+', '_', file_name)
# Remove any multiple underscores
clean_file_name = re.sub(r'_+', '_', cleaned).strip('_')
# Generate a random UUID for uniqueness
random_uuid = uuid.uuid4().hex[:6]
# Combine cleaned file name with the original extension
clean_file_path = os.path.join(os.path.dirname(file_path), clean_file_name + f"_{random_uuid}" + file_extension)
return clean_file_path
# def get_audio_file(uploaded_file):
# global base_path
# # ,device
# device = "cuda" if torch.cuda.is_available() else "cpu"
# # Detect the file type (audio/video)
# mime_type, _ = mimetypes.guess_type(uploaded_file)
# # Create the folder path to store audio files
# audio_folder = f"{base_path}/subtitle_audio"
# os.makedirs(audio_folder, exist_ok=True)
# # Initialize variable for the audio file path
# audio_file_path = ""
# if mime_type and mime_type.startswith('audio'):
# # If it's an audio file, save it as is
# audio_file_path = os.path.join(audio_folder, os.path.basename(uploaded_file))
# audio_file_path=clean_file_name(audio_file_path)
# shutil.copy(uploaded_file, audio_file_path) # Move file to audio folder
# elif mime_type and mime_type.startswith('video'):
# # If it's a video file, extract the audio
# audio_file_name = os.path.splitext(os.path.basename(uploaded_file))[0] + ".mp3"
# audio_file_path = os.path.join(audio_folder, audio_file_name)
# audio_file_path=clean_file_name(audio_file_path)
# # Extract the file extension from the uploaded file
# file_extension = os.path.splitext(uploaded_file)[1] # Includes the dot, e.g., '.mp4'
# # Generate a random UUID and create a new file name with the same extension
# random_uuid = uuid.uuid4().hex[:6]
# new_file_name = random_uuid + file_extension
# # Set the new file path in the subtitle_audio folder
# new_file_path = os.path.join(audio_folder, new_file_name)
# # Copy the original video file to the new location with the new name
# shutil.copy(uploaded_file, new_file_path)
# if device=="cuda":
# command = f"ffmpeg -hwaccel cuda -i {new_file_path} {audio_file_path} -y"
# else:
# command = f"ffmpeg -i {new_file_path} {audio_file_path} -y"
# subprocess.run(command, shell=True)
# if os.path.exists(new_file_path):
# os.remove(new_file_path)
# # Return the saved audio file path
# audio = AudioSegment.from_file(audio_file_path)
# # Get the duration in seconds
# duration_seconds = len(audio) / 1000.0 # pydub measures duration in milliseconds
# return audio_file_path,duration_seconds
def get_audio_file(uploaded_file):
global temp_folder
file_path = os.path.join(temp_folder, os.path.basename(uploaded_file))
file_path=clean_file_name(file_path)
shutil.copy(uploaded_file, file_path)
return file_path,None
def format_segments(segments):
saved_segments = list(segments)
sentence_timestamp = []
words_timestamp = []
speech_to_text = ""
for i in saved_segments:
temp_sentence_timestamp = {}
# Store sentence information in sentence_timestamp
text = i.text.strip()
sentence_id = len(sentence_timestamp) # Get the current index for the new entry
sentence_timestamp.append({
"id": sentence_id, # Use the index as the id
"text": text,
"start": i.start,
"end": i.end,
"words": [] # Initialize words as an empty list within the sentence
})
speech_to_text += text + " "
# Process each word in the sentence
for word in i.words:
word_data = {
"word": word.word.strip(),
"start": word.start,
"end": word.end
}
# Append word timestamps to the sentence's word list
sentence_timestamp[sentence_id]["words"].append(word_data)
# Optionally, add the word data to the global words_timestamp list
words_timestamp.append(word_data)
return sentence_timestamp, words_timestamp, speech_to_text
def combine_word_segments(words_timestamp, max_words_per_subtitle=8, min_silence_between_words=0.5):
before_translate = {}
id = 1
text = ""
start = None
end = None
word_count = 0
last_end_time = None
for i in words_timestamp:
try:
word = i['word']
word_start = i['start']
word_end = i['end']
# Check for sentence-ending punctuation
is_end_of_sentence = word.endswith(('.', '?', '!'))
# Check for conditions to create a new subtitle
if ((last_end_time is not None and word_start - last_end_time > min_silence_between_words)
or word_count >= max_words_per_subtitle
or is_end_of_sentence):
# Store the previous subtitle if there's any
if text:
before_translate[id] = {
"text": text,
"start": start,
"end": end
}
id += 1
# Reset for the new subtitle segment
text = word
start = word_start # Set the start time for the new subtitle
word_count = 1
else:
if word_count == 0: # First word in the subtitle
start = word_start # Ensure the start time is set
text += " " + word
word_count += 1
end = word_end # Update the end timestamp
last_end_time = word_end # Update the last end timestamp
except KeyError as e:
print(f"KeyError: {e} - Skipping word")
pass
# After the loop, make sure to add the last subtitle segment
if text:
before_translate[id] = {
"text": text,
"start": start,
"end": end
}
return before_translate
def convert_time_to_srt_format(seconds):
""" Convert seconds to SRT time format (HH:MM:SS,ms) """
hours = int(seconds // 3600)
minutes = int((seconds % 3600) // 60)
secs = int(seconds % 60)
milliseconds = int((seconds - int(seconds)) * 1000)
return f"{hours:02}:{minutes:02}:{secs:02},{milliseconds:03}"
def write_subtitles_to_file(subtitles, filename="subtitles.srt"):
folder = os.path.dirname(filename)
if not os.path.exists(folder):
os.makedirs(folder, exist_ok=True)
# Open the file with UTF-8 encoding
with open(filename, 'w', encoding='utf-8') as f:
for id, entry in subtitles.items():
# Write the subtitle index
f.write(f"{id}\n")
if entry['start'] is None or entry['end'] is None:
print(id)
# Write the start and end time in SRT format
start_time = convert_time_to_srt_format(entry['start'])
end_time = convert_time_to_srt_format(entry['end'])
f.write(f"{start_time} --> {end_time}\n")
# Write the text and speaker information
f.write(f"{entry['text']}\n\n")
def translate_subtitle(word_level_segments,Source_Language,Destination_Language):
global language_dict
store_text=""
meta_translator = pipeline(
task="translation",
model="facebook/nllb-200-distilled-600M",
torch_dtype=torch.bfloat16,device='cuda')
translated_subtitles = {}
for id, entry in word_level_segments.items():
# Access the complete text for each subtitle block
full_text = entry['text']
# Translate the entire text of the subtitle block
text_translated = meta_translator(full_text,
src_lang=language_dict[Source_Language]["meta_code"],
tgt_lang=language_dict[Destination_Language]["meta_code"])
translated_text=text_translated[0]["translation_text"]
# Reconstruct the subtitle with the translated text
translated_subtitles[id] = {
"text": translated_text,
"start": entry['start'],
"end": entry['end']
}
store_text+=translated_text.strip()+" "
del meta_translator
gc.collect()
torch.cuda.empty_cache()
return translated_subtitles,store_text
def whisper_subtitle(uploaded_file,Source_Language,Destination_Language):
global language_dict,base_path
#setup srt file names
base_name = os.path.basename(uploaded_file).rsplit('.', 1)[0][:30]
save_name = f"{base_path}/generated_subtitle/{base_name}_{Source_Language}.srt"
original_srt_name=clean_file_name(save_name)
translated_srt_name=save_name.replace(Source_Language,Destination_Language)
original_txt_name=original_srt_name.replace(".srt",".txt")
translated_txt_name=translated_srt_name.replace(".srt",".txt")
#Load model
faster_whisper_model = WhisperModel("deepdml/faster-whisper-large-v3-turbo-ct2")
audio_path,audio_duration=get_audio_file(uploaded_file)
if Source_Language=="Automatic":
segments,d = faster_whisper_model.transcribe(audio_path, word_timestamps=True)
lang_code=d.language
src_lang=get_language_name(lang_code)
else:
lang=language_dict[Source_Language]['lang_code']
segments,d = faster_whisper_model.transcribe(audio_path, word_timestamps=True,language=lang)
src_lang=Source_Language
if os.path.exists(audio_path):
os.remove(audio_path)
sentence_timestamp,words_timestamp,text=format_segments(segments)
del faster_whisper_model
gc.collect()
torch.cuda.empty_cache()
word_segments=combine_word_segments(words_timestamp, max_words_per_subtitle=8, min_silence_between_words=0.5)
write_subtitles_to_file(word_segments, filename=original_srt_name)
with open(original_txt_name, 'w', encoding='utf-8') as f1:
f1.write(text)
if Destination_Language=="Same as Source Language":
dest_lang=src_lang
else:
dest_lang=Destination_Language
if src_lang!=dest_lang:
translated_word_segments,translated_text=translate_subtitle(word_segments,src_lang,dest_lang)
write_subtitles_to_file(translated_word_segments, filename=translated_srt_name)
with open(translated_txt_name, 'w', encoding='utf-8') as f2:
f2.write(translated_text)
else:
translated_srt_name=original_srt_name
translated_txt_name=original_txt_name
return original_srt_name,translated_srt_name,original_txt_name,translated_txt_name,audio_path
#@title Using Gradio Interface
# import gradio as gr
# def subtitle_maker(Audio_or_Video_File,Source_Language,Destination_Language):
# try:
# original_srt_file,translated_srt_file,original_text_file,translated_text_file,used_audio_path=whisper_subtitle(Audio_or_Video_File,Source_Language,Destination_Language)
# except:
# original_srt_file,translated_srt_file,original_text_file,translated_text_file,used_audio_path=None,None,None,None,None
# # return original_srt_file,original_text_file,translated_srt_file,translated_text_file
# return translated_srt_file
# source_lang_list=['Automatic']
# source_lang_list.extend(available_language)
# destination_lang_list=["Same as Source Language"]
# destination_lang_list.extend(available_language)
# # demo_examples = [["/content/audio/a.mp3"]]
# gradio_inputs=[gr.File(label="Upload Audio or Video File"),
# gr.Dropdown(label="Source Language",choices=source_lang_list,value="Automatic"),
# gr.Dropdown(label="Destination Language",choices=destination_lang_list,value="Same as Source Language")]
# gradio_outputs=[gr.File(label="Original SRT File",show_label=True),
# gr.File(label="Original Text File",show_label=True),
# gr.File(label="Translated SRT File",show_label=True),
# gr.File(label="Translated Text File",show_label=True)]
# demo = gr.Interface(fn=subtitle_maker, inputs=gradio_inputs,outputs=gradio_outputs , title="Whisper-Large-V3-Turbo-Ct2 Subtitle Maker")#,examples=demo_examples)
# demo.launch(debug=True,share=True)
# import locale
# locale.getpreferredencoding = lambda: "UTF-8"
# !pwd
# %cd $install_path
# actual_duration
#@title Generate Audio File From Subtitle
from tqdm.notebook import tqdm
import subprocess
import json
import pysrt
import os
from pydub import AudioSegment
import shutil
import uuid
import re
import time
os.chdir(install_path)
Reference_Audio_File,Clone_Method,Seed,Remove_Silence_From_TTS=None,None,None,None
def your_tts(text,audio_path,language,actual_duration):
global Reference_Audio_File,Clone_Method,Seed,Remove_Silence_From_TTS
if len(text)<=135:
Split_Sentences=0
if len(text)>135:
Split_Sentences=2
# actual_duration= abs(actual_duration)
# actual_duration=None if actual_duration==0 else actual_duration
actual_duration=None
cloned_voice_path=voice_clone(Reference_Audio_File, text,remove_silence = Remove_Silence_From_TTS,fix_duration=actual_duration,chunks=Split_Sentences,exp_name=Clone_Method,progress_bar=False)
shutil.copy(cloned_voice_path,audio_path)
def get_subtitle_Dub_path(srt_file_path,Language):
file_name = os.path.splitext(os.path.basename(srt_file_path))[0]
if not os.path.exists(f"{base_path}/TTS_DUB"):
os.mkdir(f"{base_path}/TTS_DUB")
random_string = str(uuid.uuid4())[:6]
new_path=f"{base_path}/TTS_DUB/{file_name}_{random_string}.wav"
return new_path
def get_video_duration(video_path):
try:
# Run ffmpeg command to get video information in JSON format
result = subprocess.run(
['ffmpeg', '-i', video_path, '-f', 'ffmetadata', '-'],
stderr=subprocess.PIPE,
text=True
)
# Parse the duration from the stderr output
for line in result.stderr.split('\n'):
if 'Duration' in line:
duration_str = line.split('Duration: ')[1].split(',')[0]
h, m, s = duration_str.split(':')
duration = int(h) * 3600 + int(m) * 60 + float(s)
return duration
except Exception as e:
print(f"Error: {e}")
return None
# def replace_audio(video_path,audio_path):
# if not video_path.lower().endswith(".mp4"):
# return
# tts_audio = AudioSegment.from_file(dub_save_path)
# audio_duration = len(tts_audio)/1000
# video_duration=get_video_duration(video_path)
# slience_duration=video_duration-audio_duration
# audio_segment = AudioSegment.from_file(audio_path)
# slience_Segment= AudioSegment.silent(duration=slience_duration)
# marge_audio=audio_segment+slience_Segment
# marge_audio.export(f"{base_path}/new_audio.wav", format="wav")
# command=f"ffmpeg -i {video_path} -i {base_path}/new_audio.wav -map 0:v -map 1:a -c:v copy -shortest {base_path}/output.mp4 -y"
# var=os.system(command)
# if var==0: