-
Notifications
You must be signed in to change notification settings - Fork 5
/
textual_inversion_audioldm.py
1599 lines (1371 loc) · 68.5 KB
/
textual_inversion_audioldm.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
#!/usr/bin/env python
# coding=utf-8
import argparse
import logging
import math
import os
import random
import shutil
import warnings
import configparser
import ast
import contextlib
import glob
import csv
from pathlib import Path
import pandas as pd
import soundfile as sf
import numpy as np
import PIL
import torch
import torch.nn.functional as F
import torch.utils.checkpoint
import transformers
from accelerate import Accelerator
from accelerate.logging import get_logger
from accelerate.utils import ProjectConfiguration, set_seed
from huggingface_hub import create_repo, upload_folder
# TODO: remove and import from diffusers.utils when the new version of diffusers is released
from packaging import version
from PIL import Image
from torch.utils.data import Dataset
from torchvision import transforms
import torchaudio
from tqdm.auto import tqdm
from transformers import CLIPTextModel, CLIPTokenizer
from transformers import ClapTextModelWithProjection, RobertaTokenizer, RobertaTokenizerFast, SpeechT5HifiGan
import diffusers
from diffusers import (
AutoencoderKL,
DDPMScheduler,
DiffusionPipeline,
DPMSolverMultistepScheduler,
StableDiffusionPipeline,
UNet2DConditionModel,
DDIMScheduler
)
from pipeline.pipeline_audioldm import AudioLDMPipeline
from transformers import SpeechT5HifiGan
from diffusers.optimization import get_scheduler
from diffusers.utils import check_min_version, is_wandb_available
from diffusers.utils.import_utils import is_xformers_available
from audioldm.audio import wav_to_fbank, TacotronSTFT, read_wav_file
from audioldm.utils import default_audioldm_config
from scipy.io.wavfile import write
from utils.templates import imagenet_templates_small, imagenet_style_templates_small, text_editability_templates, minimal_templates, imagenet_templates_small_class
from evaluate import LAIONCLAPEvaluator
import matplotlib
matplotlib.use('Agg') # No pictures displayed
import matplotlib.pyplot as plt
import pylab
import librosa
import librosa.display
import os
if is_wandb_available():
import wandb
if version.parse(version.parse(PIL.__version__).base_version) >= version.parse("9.1.0"):
PIL_INTERPOLATION = {
"linear": PIL.Image.Resampling.BILINEAR,
"bilinear": PIL.Image.Resampling.BILINEAR,
"bicubic": PIL.Image.Resampling.BICUBIC,
"lanczos": PIL.Image.Resampling.LANCZOS,
"nearest": PIL.Image.Resampling.NEAREST,
}
else:
PIL_INTERPOLATION = {
"linear": PIL.Image.LINEAR,
"bilinear": PIL.Image.BILINEAR,
"bicubic": PIL.Image.BICUBIC,
"lanczos": PIL.Image.LANCZOS,
"nearest": PIL.Image.NEAREST,
}
# ------------------------------------------------------------------------------
# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
check_min_version("0.18.0.dev0")
logger = get_logger(__name__)
def save_model_card(repo_id: str, audios=None, base_model=str, repo_folder=None):
audio_str = ""
for i, audio in enumerate(audios):
write(os.path.join(repo_folder, f"audio_{i}.wav"),16000, audio)
audio_str += f"![aud_{i}](./audio_{i}.wav)\n"
yaml = f"""
---
license: creativeml-openrail-m
base_model: {base_model}
tags:
- stable-diffusion
- stable-diffusion-diffusers
- text-to-image
- diffusers
- textual_inversion
inference: true
---
"""
model_card = f"""
# Textual inversion text2image fine-tuning - {repo_id}
These are textual inversion adaption weights for {base_model}. You can find some example images in the following. \n
{audio_str}
"""
with open(os.path.join(repo_folder, "README.md"), "w") as f:
f.write(yaml + model_card)
def log_csv(csv_file,row):
#row is a list of strings, eg
# row = ['Jane Smith', '28', 'Designer']
# Check if the CSV file exists
if os.path.exists(csv_file):
# File exists, open in append mode
with open(csv_file, 'a', newline='') as file:
writer = csv.writer(file)
writer.writerow(row)
else:
# File does not exist, create and add line
with open(csv_file, 'w', newline='') as file:
writer = csv.writer(file)
# header = ['Name', 'Age', 'Occupation']
# writer.writerow(header)
writer.writerow(row)
def create_mixture(waveform1, waveform2, snr):
min_length = min(waveform1.shape[1], waveform2.shape[1])
waveform1 = waveform1[:, :min_length]
waveform2 = waveform2[:, :min_length]
# Calculate the power of each waveform
power1 = torch.mean(waveform1 ** 2)
power2 = torch.mean(waveform2 ** 2)
# Calculate the desired power ratio based on SNR (Signal-to-Noise Ratio)
desired_snr = 10 ** (-snr / 10)
scale_factor = torch.sqrt(desired_snr * power1 / power2)
# Scale the second waveform to achieve the desired SNR
scaled_waveform2 = waveform2 * scale_factor
mixture = waveform1 + scaled_waveform2
return mixture.numpy()
def log_validation(text_encoder, tokenizer, unet, vae, args, accelerator, weight_dtype, global_step,vocoder,concept_audio_dir,placeholder_token_ids, validate_experiments=False):
logger.info(
f"Running validation... \n Generating {args.num_validation_audio_files} audio files with prompt:"
f" {args.validation_prompt}."
)
# create pipeline (note: unet and vae are loaded again in float32)
pipeline = AudioLDMPipeline.from_pretrained(
args.pretrained_model_name_or_path,
text_encoder=accelerator.unwrap_model(text_encoder),
tokenizer=tokenizer,
unet=unet,
vae=vae,
safety_checker=None,
revision=args.revision,
torch_dtype=weight_dtype,
)
# pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
pipeline = pipeline.to(accelerator.device)
pipeline.set_progress_bar_config(disable=True)
# run inference
generator = None if args.seed is None else torch.Generator(device=accelerator.device).manual_seed(args.seed)
audios = []
for _ in range(args.num_validation_audio_files):
with torch.autocast("cuda"):
audio_gen = pipeline(args.validation_prompt, num_inference_steps=50,audio_length_in_s=10, generator=generator).audios[0]
audios.append(audio_gen)
for tracker in accelerator.trackers:
if tracker.name == "tensorboard":
# np_images = np.stack([np.asarray(img) for img in images])
for i, audio in enumerate(audios):
tracker.writer.add_audio(f"validation_{i}", audio, global_step, sample_rate=16000)
if tracker.name == "wandb":
# tracker.log(
# {
# "validation": [
# wandb.Image(image, caption=f"{i}: {args.validation_prompt}") for i, image in enumerate(images)
# ]
# }
# )
raise NotImplementedError("Wandb not implemented yet for audio")
if validate_experiments:
# TODO check if audio length should be set, for now setting to training length
audios_rec = pipeline(args.validation_prompt, num_inference_steps=50,audio_length_in_s=10.24, generator=generator,num_waveforms_per_prompt=10).audios
# print(audios_rec.shape)
# audios_rec = np.concatenate((audios_rec, pipeline(args.validation_prompt, num_inference_steps=50, generator=generator,num_waveforms_per_prompt=20).audios))
# print("audios_rec shape: {}".format(audios_rec.shape))
val_audio_dir = os.path.join(args.output_dir, "reconstruction_audio_{}".format(global_step))
os.makedirs(val_audio_dir, exist_ok=True)
for i, audio in enumerate(audios_rec):
write(os.path.join(val_audio_dir, f"{'_'.join(args.validation_prompt.split(' '))}_{i}.wav"),16000, audio)
print("loading clap evaluator")
with contextlib.redirect_stdout(None):
evaluator = LAIONCLAPEvaluator(device=accelerator.device)
reconstruction_score=evaluator.audio_to_audio_similarity(concept_audio_dir, val_audio_dir)
print("Reconstruction score: {}".format(reconstruction_score))
print(type(reconstruction_score))
accelerator.log({"reconstruction_score": reconstruction_score.item()},step=global_step)
reconstruction_csv_path=os.path.join(args.output_dir, "reconstruction_score.csv")
log_csv(reconstruction_csv_path,[global_step,reconstruction_score.item()])
# accelerator.log({"reconstruction_scored": np.float16(reconstruction_score)},step=global_step)
del audios_rec
# val_audio_dir = os.path.join(args.output_dir, "editability_audio_{}".format(global_step))
# os.makedirs(val_audio_dir, exist_ok=True)
# for prompt in text_editability_templates:
# prompt=" ".join(prompt.format(("".join(tokenizer.convert_ids_to_tokens(placeholder_token_ids)))).split())
# audio = pipeline(prompt, num_inference_steps=50, generator=generator,num_waveforms_per_prompt=1).audios[0]
# write(os.path.join(val_audio_dir, f"{'_'.join(prompt.split(' '))}.wav"),16000, audio)
# text_score=evaluator.text_to_audio_similarity(val_audio_dir)
# print("Text score: {}".format(text_score))
# accelerator.log({"text_score": text_score}, step=global_step)
del evaluator
del pipeline
torch.cuda.empty_cache()
return audios
def save_progress(text_encoder, placeholder_token_ids, accelerator, args, save_path):
logger.info("Saving embeddings")
learned_embeds = (
accelerator.unwrap_model(text_encoder)
.get_input_embeddings()
.weight[min(placeholder_token_ids) : max(placeholder_token_ids) + 1]
)
learned_embeds_dict = {args.placeholder_token: learned_embeds.detach().cpu()}
torch.save(learned_embeds_dict, save_path)
def parse_args():
parser = argparse.ArgumentParser(description="Simple example of a training script.")
parser.add_argument("--config", type=str, default=None, help="Path to .ini file.")
parser.add_argument(
"--save_steps",
type=int,
default=500,
help="Save learned_embeds.bin every X updates steps.",
)
parser.add_argument(
"--save_as_full_pipeline",
action="store_true",
help="Save the complete stable diffusion pipeline.",
)
parser.add_argument(
"--num_vectors",
type=int,
default=1,
help="How many textual inversion vectors shall be used to learn the concept.",
)
parser.add_argument(
"--pretrained_model_name_or_path",
type=str,
default=None,
required=False,
help="Path to pretrained model or model identifier from huggingface.co/models.",
)
parser.add_argument(
"--revision",
type=str,
default=None,
required=False,
help="Revision of pretrained model identifier from huggingface.co/models.",
)
parser.add_argument(
"--tokenizer_name",
type=str,
default=None,
help="Pretrained tokenizer name or path if not the same as model_name",
)
parser.add_argument(
"--train_data_dir", type=str, default=None, required=False, help="A folder containing the training data."
)
parser.add_argument(
"--file_list", type=str, default=None, help="Path to a csv file containing which files to train on from the training data directory."
)
parser.add_argument(
"--placeholder_token",
type=str,
default=None,
required=False,
help="A token to use as a placeholder for the concept.",
)
parser.add_argument("--initializer", type=str, default="random_token",choices=["random_token","random_tokens","multitoken_word","saved_embedding","mean","multiresolution","mean_word"], help="How to initialize the placeholder.")
parser.add_argument(
"--initializer_token", type=str, default=None, required=False, help="A token to use as initializer word."
)
parser.add_argument("--learnable_property", type=str, default="object", help="Choose between 'object' and 'style'")
parser.add_argument("--object_class", type=str, default=None, help="Choose a class to learn, works with learnable property 'object_class'")
parser.add_argument("--repeats", type=int, default=100, help="How many times to repeat the training data.")
parser.add_argument(
"--output_dir",
type=str,
default="audio-inversion-model",
help="The output directory where the model predictions and checkpoints will be written.",
)
parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
#todo:change resolution
# parser.add_argument(
# "--resolution",
# type=int,
# default=512,
# help=(
# "The resolution for input images, all the images in the train/validation dataset will be resized to this"
# " resolution"
# ),
# )
parser.add_argument("--sample_rate", type=int, default=16000, help="Sample rate for audio.")
parser.add_argument("--duration", type=float, default=10.0, help="Duration of audio.")
# parser.add_argument(
# "--center_crop", action="store_true", help="Whether to center crop images before resizing to resolution."
# )
parser.add_argument(
"--train_batch_size", type=int, default=16, help="Batch size (per device) for the training dataloader."
)
parser.add_argument("--num_train_epochs", type=int, default=100)
parser.add_argument(
"--max_train_steps",
type=int,
default=5000,
help="Total number of training steps to perform. If provided, overrides num_train_epochs.",
)
parser.add_argument(
"--gradient_accumulation_steps",
type=int,
default=1,
help="Number of updates steps to accumulate before performing a backward/update pass.",
)
parser.add_argument(
"--gradient_checkpointing",
action="store_true",
help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
)
parser.add_argument(
"--learning_rate",
type=float,
default=1e-4,
help="Initial learning rate (after the potential warmup period) to use.",
)
parser.add_argument(
"--scale_lr",
action="store_true",
default=False,
help="Scale the learning rate by the number of GPUs, gradient accumulation steps, and batch size.",
)
parser.add_argument(
"--lr_scheduler",
type=str,
default="constant",
help=(
'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
' "constant", "constant_with_warmup"]'
),
)
parser.add_argument(
"--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
)
parser.add_argument(
"--lr_num_cycles",
type=int,
default=1,
help="Number of hard resets of the lr in cosine_with_restarts scheduler.",
)
parser.add_argument(
"--dataloader_num_workers",
type=int,
default=0,
help=(
"Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
),
)
parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
parser.add_argument("--adam_weight_decay", type=float, default=1e-2, help="Weight decay to use.")
parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
parser.add_argument(
"--hub_model_id",
type=str,
default=None,
help="The name of the repository to keep in sync with the local `output_dir`.",
)
parser.add_argument(
"--logging_dir",
type=str,
default="logs",
help=(
"[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
" *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
),
)
parser.add_argument(
"--mixed_precision",
type=str,
default="no",
choices=["no", "fp16", "bf16"],
help=(
"Whether to use mixed precision. Choose"
"between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
"and an Nvidia Ampere GPU."
),
)
parser.add_argument(
"--allow_tf32",
action="store_true",
help=(
"Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
" https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
),
)
parser.add_argument(
"--report_to",
type=str,
default="tensorboard",
help=(
'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
),
)
parser.add_argument(
"--validation_prompt",
type=str,
default=None,
help="A prompt that is used during validation to verify that the model is learning.",
)
parser.add_argument(
"--num_validation_audio_files",
type=int,
default=4,
help="Number of audio files that should be generated during validation with `validation_prompt`.",
)
parser.add_argument(
"--validation_steps",
type=int,
default=500,
help=(
"Run validation every X steps. Validation consists of running the prompt"
" `args.validation_prompt` multiple times: `args.num_validation_images`"
" and logging the images."
),
)
parser.add_argument(
"--validation_epochs",
type=int,
default=None,
help=(
"Deprecated in favor of validation_steps. Run validation every X epochs. Validation consists of running the prompt"
" `args.validation_prompt` multiple times: `args.num_validation_images`"
" and logging the images."
),
)
parser.add_argument("--validate_experiments", action="store_true", help="Whether to validate experiments.")
parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
parser.add_argument(
"--checkpointing_steps",
type=int,
default=500,
help=(
"Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
" training using `--resume_from_checkpoint`."
),
)
parser.add_argument(
"--checkpoints_total_limit",
type=int,
default=None,
help=("Max number of checkpoints to store."),
)
parser.add_argument(
"--resume_from_checkpoint",
type=str,
default=None,
help=(
"Whether training should be resumed from a previous checkpoint. Use a path saved by"
' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
),
)
parser.add_argument(
"--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
)
parser.add_argument("--save_concept_audio", action="store_true",default=True, help="Whether or not to save concept audio.")
parser.add_argument("--augment_data", action="store_true",default=False, help="Whether or not to augment the training data")
parser.add_argument("--num_scales", type=int, default=None, help="Number of scales for multi-resolution tarining")
parser.add_argument("--mix_data", type=str,default=None, help="If a path to an dir containing background audios is specified performs mixture training")
parser.add_argument(
"--snr",
type=int,
default=20,
help="In mixture training specify SNR of",
)
parser.add_argument(
"--num_audio_files_to_train",
type=int,
default=None,
help="Number of files to use for training if None will use all files in training dir",
)
def read_args_from_config(filename):
config = configparser.ConfigParser()
config.read(filename)
args = dict(config["Arguments"])
# Convert the values to the appropriate data types
for key, value in args.items():
try:
args[key] = ast.literal_eval(value)
except (ValueError, SyntaxError):
pass # If the value cannot be evaluated, keep it as a string
return args
cli_args, _ = parser.parse_known_args()
if cli_args.config:
print("Reading arguments from config file")
config_args = read_args_from_config(cli_args.config)
# Update the argparse namespace with config_args
for key, value in config_args.items():
setattr(cli_args, key, value)
args = cli_args
# args = parser.parse_args()
env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
if env_local_rank != -1 and env_local_rank != args.local_rank:
args.local_rank = env_local_rank
if args.train_data_dir is None:
raise ValueError("You must specify a train data directory.")
return args
from audioldm.audio.tools import get_mel_from_wav, _pad_spec, normalize_wav, pad_wav
def read_wav_file(filename, segment_length, augment_data=False):
# waveform, sr = librosa.load(filename, sr=None, mono=True) # 4 times slower
waveform, sr = torchaudio.load(filename) # Faster!!!
waveform = torchaudio.functional.resample(waveform, orig_freq=sr, new_freq=16000)
if augment_data:
waveform = augment_audio(
waveform,
sr,
p=0.8,
noise=True,
reverb=True,
low_pass=True,
pitch_shift=True,
delay=True)
waveform = waveform.numpy()[0, ...]
waveform = normalize_wav(waveform)
waveform = waveform[None, ...]
waveform = pad_wav(waveform, segment_length)
waveform = waveform / np.max(np.abs(waveform))
waveform = 0.5 * waveform
return waveform
def wav_to_fbank(
filename,
target_length=1024,
fn_STFT=None,
augment_data=False,
mix_data=False,
snr=None
):
assert fn_STFT is not None
# mixup
if mix_data:
assert snr is not None, "You specified mixed training but didn't provide SNR!"
background_file_paths = [os.path.join(mix_data, p) for p in os.listdir(mix_data)]
background_file_path = random.sample(background_file_paths,1)[0]
waveform = read_wav_file(filename, target_length * 160, augment_data=augment_data)
background = read_wav_file(background_file_path, target_length * 160)
waveform = create_mixture(torch.tensor(waveform), torch.tensor(background), snr)
else:
waveform = read_wav_file(filename, target_length * 160, augment_data=augment_data) # hop size is 160
waveform = waveform[0, ...]
waveform = torch.FloatTensor(waveform)
fbank, log_magnitudes_stft, energy = get_mel_from_wav(waveform, fn_STFT)
fbank = torch.FloatTensor(fbank.T)
log_magnitudes_stft = torch.FloatTensor(log_magnitudes_stft.T)
fbank, log_magnitudes_stft = _pad_spec(fbank, target_length), _pad_spec(
log_magnitudes_stft, target_length
)
return fbank, log_magnitudes_stft, waveform
def wav_to_mel(
original_audio_file_path,
duration,
augment_data=False,
mix_data=False,
snr=None
):
config=default_audioldm_config()
fn_STFT = TacotronSTFT(
config["preprocessing"]["stft"]["filter_length"],
config["preprocessing"]["stft"]["hop_length"],
config["preprocessing"]["stft"]["win_length"],
config["preprocessing"]["mel"]["n_mel_channels"],
config["preprocessing"]["audio"]["sampling_rate"],
config["preprocessing"]["mel"]["mel_fmin"],
config["preprocessing"]["mel"]["mel_fmax"],
)
mel, _, _ = wav_to_fbank(
original_audio_file_path,
target_length=int(duration * 102.4),
fn_STFT=fn_STFT,
augment_data=augment_data,
mix_data=mix_data,
snr=snr
)
mel = mel.unsqueeze(0)
# mel = repeat(mel, "1 ... -> b ...", b=batchsize)
if augment_data:
mel=mel.unsqueeze(0)
mel = augment_spectrogram(mel)
mel = mel.squeeze(0)
return mel
class AudioInversionDataset(Dataset):
def __init__(
self,
data_root,
tokenizer,
text_encoder,
device,
audioldmpipeline,
learnable_property="object", # [object, style, minimal]
sample_rate=16000,
duration=2.0,
repeats=100,
set="train",
placeholder_token="*",
file_list=None,
object_class=None,
augment_data=False,
multiresolution=False,
train_timesteps=None,
num_scales=None,
mix_data=False,
snr=None,
num_files_to_train=None
):
self.data_root = data_root
self.tokenizer = tokenizer
self.learnable_property = learnable_property
self.sample_rate = sample_rate
self.duration = duration
self.placeholder_token = placeholder_token
self.audioldmpipeline = audioldmpipeline
self.text_encoder = text_encoder
self.num_files_to_train = num_files_to_train
if file_list is not None:
file_list=list(pd.read_csv(file_list, header=None)[0])
self.audio_files = [os.path.join(self.data_root, file_path) for file_path in file_list]
else:
self.audio_files = [os.path.join(self.data_root, file_path) for file_path in os.listdir(self.data_root) if file_path.endswith(".wav")]
if self.num_files_to_train:
self.audio_files = sorted(self.audio_files)[:self.num_files_to_train]
self.num_files = len(self.audio_files)
self._length = self.num_files
self.device = device
self.augment_data = augment_data
self.multiresolution = multiresolution
self.num_train_timesteps = train_timesteps
self.num_scales = num_scales
self.mix_data = mix_data
self.snr = snr
if self.learnable_property == "object":
self.templates = imagenet_templates_small
elif self.learnable_property == "style":
self.templates = imagenet_style_templates_small
elif self.learnable_property == "minimal":
self.templates = minimal_templates
elif self.learnable_property == "object_class":
self.templates = imagenet_templates_small_class
self.object_class = object_class
if set == "train":
self._length = self.num_files * repeats
def __len__(self):
return self._length
def __getitem__(self, i):
example = {}
audio_file = self.audio_files[i % self.num_files]
example["mel"]=wav_to_mel(
audio_file,
self.duration,
augment_data=self.augment_data,
mix_data=self.mix_data,
snr=self.snr
)
# Sample a random timestep for diffusion
timestep = torch.randint(
0, self.num_train_timesteps, (1,)
).long()
if self.multiresolution:
scale_index = int((timestep / self.num_train_timesteps) * self.num_scales)
placeholder_string = f"<{self.placeholder_token.split('|')[0][1:]}|{scale_index}|>"
text = random.choice(self.templates).format(placeholder_string)
else:
placeholder_string = self.placeholder_token
if self.learnable_property == "object_class":
text = random.choice(self.templates).format(placeholder_string, self.object_class)
else:
text = random.choice(self.templates).format(placeholder_string)
# text_inputs = self.tokenizer(
# text,
# padding="max_length",
# max_length=self.tokenizer.model_max_length,
# truncation=True,
# return_tensors="pt",
# )
# text_input_ids = text_inputs.input_ids
# attention_mask = text_inputs.attention_mask
# prompt_embeds = self.text_encoder(
# text_input_ids.to(self.device),
# attention_mask=attention_mask.to(self.device),
# )
# prompt_embeds = prompt_embeds.text_embeds
# prompt_embeds = F.normalize(prompt_embeds, dim=-1)
# prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=self.device)
# example["prompt_embeds"] = prompt_embeds
#alternatively, we can get prompt embeds by the following from the audioldmpipeline
#there we can also do classifier free guidance during training if we want
example["prompt_embeds"] = self.audioldmpipeline._encode_prompt(
prompt=text,
device=self.device,
num_waveforms_per_prompt=1,
do_classifier_free_guidance=False
)
return example, timestep
def main():
args = parse_args()
logging_dir = os.path.join(args.output_dir, args.logging_dir)
accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
accelerator = Accelerator(
gradient_accumulation_steps=args.gradient_accumulation_steps,
mixed_precision=args.mixed_precision,
log_with=args.report_to,
project_config=accelerator_project_config,
)
if args.report_to == "wandb":
if not is_wandb_available():
raise ImportError("Make sure to install wandb if you want to use it for logging during training.")
# Make one log on every process with the configuration for debugging.
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
logger.info(accelerator.state, main_process_only=False)
if accelerator.is_local_main_process:
transformers.utils.logging.set_verbosity_warning()
diffusers.utils.logging.set_verbosity_info()
else:
transformers.utils.logging.set_verbosity_error()
diffusers.utils.logging.set_verbosity_error()
# If passed along, set the training seed now.
if args.seed is not None:
set_seed(args.seed)
# Handle the repository creation
if accelerator.is_main_process:
if args.output_dir is not None:
os.makedirs(args.output_dir, exist_ok=True)
if args.push_to_hub:
repo_id = create_repo(
repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
).repo_id
# Load tokenizer
if args.tokenizer_name:
tokenizer = RobertaTokenizerFast.from_pretrained(args.tokenizer_name)
elif args.pretrained_model_name_or_path:
tokenizer = RobertaTokenizerFast.from_pretrained(args.pretrained_model_name_or_path, subfolder="tokenizer")
# Load scheduler and models
noise_scheduler = DDIMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
text_encoder = ClapTextModelWithProjection.from_pretrained(
args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision
)
vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder="vae", revision=args.revision)
unet = UNet2DConditionModel.from_pretrained(
args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision
)
vocoder = SpeechT5HifiGan.from_pretrained(args.pretrained_model_name_or_path, subfolder="vocoder", revision=args.revision
)
audioldmpipeline=AudioLDMPipeline(
text_encoder=text_encoder,
vae=vae,
unet=unet,
vocoder=vocoder,
scheduler=noise_scheduler,
tokenizer=tokenizer
).to(accelerator.device)
# Add the placeholder token in tokenizer
placeholder_tokens = [args.placeholder_token]
if args.num_vectors < 1:
raise ValueError(f"--num_vectors has to be larger or equal to 1, but is {args.num_vectors}")
init_embeddings_save_path=""
if args.initializer=="mean_word":
print("Starting from the mean embedding of the placeholder token subwords.")
token_ids = tokenizer.encode(args.initializer_token, add_special_tokens=False)
if len(token_ids) > 1:
token_embeds = text_encoder.get_input_embeddings().weight.data
init_token_embed=token_embeds[token_ids].mean(dim=0)
if args.num_vectors > 1:
print("Adding {} tokens to the tokenizer.".format(str(args.num_vectors)))
# add dummy tokens for multi-vector
additional_tokens = []
for i in range(1, args.num_vectors):
additional_tokens.append(f"{args.placeholder_token}_{i}")
placeholder_tokens += additional_tokens
# the following is wrong
num_added_tokens = tokenizer.add_tokens(placeholder_tokens)
if num_added_tokens != args.num_vectors:
raise ValueError(
f"The tokenizer already contains the token {args.placeholder_token}. Please pass a different"
" `placeholder_token` that is not already in the tokenizer."
)
# initializer_token_id = token_ids[0]
placeholder_token_ids = tokenizer.convert_tokens_to_ids(placeholder_tokens)
print("placeholder_token_ids: ", placeholder_token_ids)
# Resize the token embeddings as we are adding new special tokens to the tokenizer
text_encoder.resize_token_embeddings(len(tokenizer))
# Initialise the newly added placeholder token with a random embedding, and save the embedding to file
token_embeds = text_encoder.get_input_embeddings().weight.data
with torch.no_grad():
# random_embedding = torch.rand(768).to(token_embeds.device)
#save embedding to file
init_embed_dict = {}
init_embeddings_save_path=os.path.join(args.output_dir, "initial_embeds.bin")
for token_id in placeholder_token_ids:
init_embed_dict [tokenizer.convert_ids_to_tokens(token_id)] = init_token_embed.clone().detach().cpu().unsqueeze(0)
# token_embeds[token_id] = token_embeds[initializer_token_id].clone()
token_embeds[token_id] = init_token_embed.clone()
if not args.resume_from_checkpoint:
torch.save(init_embed_dict, init_embeddings_save_path)
elif args.initializer=="random_token":
print("Starting from a random embedding. If num_tokens>1, all tokens start with the same embedding.")
if args.num_vectors > 1:
print("Adding {} tokens to the tokenizer.".format(str(args.num_vectors)))
# add dummy tokens for multi-vector
additional_tokens = []
for i in range(1, args.num_vectors):
additional_tokens.append(f"{args.placeholder_token}_{i}")
placeholder_tokens += additional_tokens
# the following is wrong
num_added_tokens = tokenizer.add_tokens(placeholder_tokens)
if num_added_tokens != args.num_vectors:
raise ValueError(
f"The tokenizer already contains the token {args.placeholder_token}. Please pass a different"
" `placeholder_token` that is not already in the tokenizer."
)
# initializer_token_id = token_ids[0]
placeholder_token_ids = tokenizer.convert_tokens_to_ids(placeholder_tokens)
print("placeholder_token_ids: ", placeholder_token_ids)
# Resize the token embeddings as we are adding new special tokens to the tokenizer
text_encoder.resize_token_embeddings(len(tokenizer))
# Initialise the newly added placeholder token with a random embedding, and save the embedding to file
token_embeds = text_encoder.get_input_embeddings().weight.data
with torch.no_grad():
random_embedding = torch.rand(768).to(token_embeds.device)
#save embedding to file
init_embed_dict = {}
init_embeddings_save_path=os.path.join(args.output_dir, "initial_embeds.bin")
for token_id in placeholder_token_ids:
init_embed_dict [tokenizer.convert_ids_to_tokens(token_id)] = random_embedding.clone().detach().cpu().unsqueeze(0)
# token_embeds[token_id] = token_embeds[initializer_token_id].clone()
token_embeds[token_id] = random_embedding.clone()
if not args.resume_from_checkpoint:
torch.save(init_embed_dict, init_embeddings_save_path)
elif args.initializer=="multiresolution":
assert args.num_vectors == 1, "Multiresolution training works with 1 vector per resolution for now"
placeholder_tokens = []
for scale_index in range(args.num_scales):
placeholder_string = f"<{args.placeholder_token}|{scale_index}|>"
placeholder_tokens.append(placeholder_string)
increment = tokenizer.add_tokens(placeholder_tokens)
print("incriment:", increment)
# if increment == 0:
# raise ValueError(
# f"The tokenizer already contains the token {placeholder_string}. Please pass a different"
# " `placeholder_token` that is not already in the tokenizer."
# )
placeholder_token_ids = tokenizer.convert_tokens_to_ids(placeholder_tokens)
print("placeholder_token_ids: ", placeholder_token_ids)
# Resize the token embeddings as we are adding new special tokens to the tokenizer
text_encoder.resize_token_embeddings(len(tokenizer))
# Initialise the newly added placeholder token with a random embedding, and save the embedding to file
token_embeds = text_encoder.get_input_embeddings().weight.data
with torch.no_grad():
random_embedding = torch.rand(768).to(token_embeds.device)
#save embedding to file
init_embed_dict = {}
init_embeddings_save_path=os.path.join(args.output_dir, "initial_embeds.bin")