Spaces:
Paused
Paused
Update generate_multitalk.py
Browse files- generate_multitalk.py +39 -40
generate_multitalk.py
CHANGED
@@ -23,7 +23,6 @@ from wan.utils.utils import cache_image, cache_video, str2bool
|
|
23 |
from wan.utils.multitalk_utils import save_video_ffmpeg
|
24 |
|
25 |
from transformers import Wav2Vec2FeatureExtractor
|
26 |
-
from transformers import Wav2Vec2ForCTC
|
27 |
from src.audio_analysis.wav2vec2 import Wav2Vec2Model
|
28 |
|
29 |
import librosa
|
@@ -215,7 +214,7 @@ def _parse_args():
|
|
215 |
return args
|
216 |
|
217 |
def custom_init(device, wav2vec):
|
218 |
-
audio_encoder = Wav2Vec2Model.from_pretrained(
|
219 |
audio_encoder.freeze_feature_extractor()
|
220 |
wav2vec_feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(wav2vec, local_files_only=True)
|
221 |
return wav2vec_feature_extractor, audio_encoder
|
@@ -373,50 +372,50 @@ def generate(args):
|
|
373 |
|
374 |
assert args.task == "multitalk-14B", 'You should choose multitalk in args.task.'
|
375 |
|
376 |
-
#
|
377 |
-
input_data = None
|
378 |
-
|
379 |
-
# Let only the main process prepare the data
|
380 |
if rank == 0:
|
381 |
with open(args.input_json, 'r', encoding='utf-8') as f:
|
382 |
input_data = json.load(f)
|
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 |
if dist.is_initialized():
|
413 |
-
objects_to_broadcast = [input_data] if rank == 0 else [None]
|
414 |
-
dist.broadcast_object_list(objects_to_broadcast, src=0)
|
415 |
-
input_data = objects_to_broadcast[0]
|
416 |
-
|
417 |
-
# Wait for all file I/O to be complete before proceeding
|
418 |
dist.barrier()
|
419 |
|
|
|
|
|
|
|
|
|
420 |
logging.info("Creating MultiTalk pipeline.")
|
421 |
wan_i2v = wan.MultiTalkPipeline(
|
422 |
config=cfg,
|
@@ -461,7 +460,7 @@ def generate(args):
|
|
461 |
args.save_file = f"{args.task}_{args.size.replace('*','x') if sys.platform=='win32' else args.size}_{args.ulysses_size}_{args.ring_size}_{formatted_prompt}_{formatted_time}"
|
462 |
|
463 |
logging.info(f"Saving generated video to {args.save_file}.mp4")
|
464 |
-
|
465 |
|
466 |
logging.info("Finished.")
|
467 |
|
|
|
23 |
from wan.utils.multitalk_utils import save_video_ffmpeg
|
24 |
|
25 |
from transformers import Wav2Vec2FeatureExtractor
|
|
|
26 |
from src.audio_analysis.wav2vec2 import Wav2Vec2Model
|
27 |
|
28 |
import librosa
|
|
|
214 |
return args
|
215 |
|
216 |
def custom_init(device, wav2vec):
|
217 |
+
audio_encoder = Wav2Vec2Model.from_pretrained(wav2vec, attn_implementation="eager").to(device)
|
218 |
audio_encoder.freeze_feature_extractor()
|
219 |
wav2vec_feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(wav2vec, local_files_only=True)
|
220 |
return wav2vec_feature_extractor, audio_encoder
|
|
|
372 |
|
373 |
assert args.task == "multitalk-14B", 'You should choose multitalk in args.task.'
|
374 |
|
375 |
+
# Let only the main process (rank 0) prepare the audio embeddings and overwrite the input JSON file.
|
|
|
|
|
|
|
376 |
if rank == 0:
|
377 |
with open(args.input_json, 'r', encoding='utf-8') as f:
|
378 |
input_data = json.load(f)
|
379 |
|
380 |
+
wav2vec_feature_extractor, audio_encoder = custom_init('cpu', args.wav2vec_dir)
|
381 |
+
args.audio_save_dir = os.path.join(args.audio_save_dir, input_data['cond_image'].split('/')[-1].split('.')[0])
|
382 |
+
os.makedirs(args.audio_save_dir, exist_ok=True)
|
383 |
+
|
384 |
+
if len(input_data['cond_audio']) == 2:
|
385 |
+
new_human_speech1, new_human_speech2, sum_human_speechs = audio_prepare_multi(input_data['cond_audio']['person1'], input_data['cond_audio']['person2'], input_data['audio_type'])
|
386 |
+
audio_embedding_1 = get_embedding(new_human_speech1, wav2vec_feature_extractor, audio_encoder)
|
387 |
+
audio_embedding_2 = get_embedding(new_human_speech2, wav2vec_feature_extractor, audio_encoder)
|
388 |
+
emb1_path = os.path.join(args.audio_save_dir, '1.pt')
|
389 |
+
emb2_path = os.path.join(args.audio_save_dir, '2.pt')
|
390 |
+
sum_audio = os.path.join(args.audio_save_dir, 'sum.wav')
|
391 |
+
sf.write(sum_audio, sum_human_speechs, 16000)
|
392 |
+
torch.save(audio_embedding_1, emb1_path)
|
393 |
+
torch.save(audio_embedding_2, emb2_path)
|
394 |
+
input_data['cond_audio']['person1'] = emb1_path
|
395 |
+
input_data['cond_audio']['person2'] = emb2_path
|
396 |
+
input_data['video_audio'] = sum_audio
|
397 |
+
elif len(input_data['cond_audio']) == 1:
|
398 |
+
human_speech = audio_prepare_single(input_data['cond_audio']['person1'])
|
399 |
+
audio_embedding = get_embedding(human_speech, wav2vec_feature_extractor, audio_encoder)
|
400 |
+
emb_path = os.path.join(args.audio_save_dir, '1.pt')
|
401 |
+
sum_audio = os.path.join(args.audio_save_dir, 'sum.wav')
|
402 |
+
sf.write(sum_audio, human_speech, 16000)
|
403 |
+
torch.save(audio_embedding, emb_path)
|
404 |
+
input_data['cond_audio']['person1'] = emb_path
|
405 |
+
input_data['video_audio'] = sum_audio
|
406 |
+
|
407 |
+
# Overwrite the temporary JSON file with the updated paths
|
408 |
+
with open(args.input_json, 'w', encoding='utf-8') as f:
|
409 |
+
json.dump(input_data, f, indent=4)
|
410 |
+
|
411 |
+
# Barrier to ensure rank 0 has finished writing all files (embeddings AND the json)
|
412 |
if dist.is_initialized():
|
|
|
|
|
|
|
|
|
|
|
413 |
dist.barrier()
|
414 |
|
415 |
+
# Now, ALL processes read the (now updated) JSON file, ensuring data consistency.
|
416 |
+
with open(args.input_json, 'r', encoding='utf-8') as f:
|
417 |
+
input_data = json.load(f)
|
418 |
+
|
419 |
logging.info("Creating MultiTalk pipeline.")
|
420 |
wan_i2v = wan.MultiTalkPipeline(
|
421 |
config=cfg,
|
|
|
460 |
args.save_file = f"{args.task}_{args.size.replace('*','x') if sys.platform=='win32' else args.size}_{args.ulysses_size}_{args.ring_size}_{formatted_prompt}_{formatted_time}"
|
461 |
|
462 |
logging.info(f"Saving generated video to {args.save_file}.mp4")
|
463 |
+
save__video_ffmpeg(video, args.save_file, [input_data['video_audio']])
|
464 |
|
465 |
logging.info("Finished.")
|
466 |
|