File size: 39,959 Bytes
672187c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "75b58048-7d14-4fc6-8085-1fc08c81b4a6",
   "metadata": {},
   "source": [
    "# Fine-Tune Whisper With 🤗 Transformers and Streaming Mode"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fbfa8ad5-4cdc-4512-9058-836cbbf65e1a",
   "metadata": {},
   "source": [
    "In this Colab, we present a step-by-step guide on fine-tuning Whisper with Hugging Face 🤗 Transformers on 400 hours of speech data! Using streaming mode, we'll show how you can train a speech recongition model on any dataset, irrespective of size. With streaming mode, storage requirements are no longer a consideration: you can train a model on whatever dataset you want, even if it's download size exceeds your devices disk space. How can this be possible? It simply seems too good to be true! Well, rest assured it's not 😉 Carry on reading to find out more."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "afe0d503-ae4e-4aa7-9af4-dbcba52db41e",
   "metadata": {},
   "source": [
    "## Introduction"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9ae91ed4-9c3e-4ade-938e-f4c2dcfbfdc0",
   "metadata": {},
   "source": [
    "Speech recognition datasets are large. A typical speech dataset consists of approximately 100 hours of audio-transcription data, requiring upwards of 130GB of storage space for download and preparation. For most ASR researchers, this is already at the upper limit of what is feasible for disk space. So what happens when we want to train on a larger dataset? The full [LibriSpeech](https://huggingface.co/datasets/librispeech_asr) dataset consists of 960 hours of audio data. Kensho's [SPGISpeech](https://huggingface.co/datasets/kensho/spgispeech) contains 5,000 hours of audio data. ML Commons [People's Speech](https://huggingface.co/datasets/MLCommons/peoples_speech) contains **30,000+** hours of audio data! Do we need to bite the bullet and buy additional storage? Or is there a way we can train on all of these datasets with no disk drive requirements?\n",
    "\n",
    "When training machine learning systems, we rarely use the entire dataset at once. We typically _batch_ our data into smaller subsets of data, and pass these incrementally through our training pipeline. This is because we train our system on an accelerator device, such as a GPU or TPU, which has a memory limit typically around 16GB. We have to fit our model, optimiser and training data all on the same accelerator device, so we usually have to divide the dataset up into smaller batches and move them from the CPU to the GPU when required.\n",
    "\n",
    "Consequently, we don't require the entire dataset to be downloaded at once; we simply need the batch of data that we pass to our model at any one go. We can leverage this principle of partial dataset loading when preparing our dataset: rather than downloading the entire dataset at the start, we can load each piece of data as and when we need it. For each batch, we load the relevant data from a remote server and pass it through the training pipeline. For the next batch, we load the next items and again pass them through the training pipeline. At no point do we have to save data to our disk drive, we simply load them in memory and use them in our pipeline. In doing so, we only ever need as much memory as each individual batch requires.\n",
    "\n",
    "This is analogous to downloading a TV show versus streaming it 📺 When we download a TV show, we download the entire video offline and save it to our disk. Compare this to when we stream a TV show. Here, we don't download any part of the video to memory, but iterate over the video file and load each part in real-time as required. It's this same principle that we can apply to our ML training pipeline! We want to iterate over the dataset and load each sample of data as required.\n",
    "\n",
    "While the principle of partial dataset loading sounds ideal, it also seems **pretty** difficult to do. Luckily for us, 🤗 Datasets allows us to do this with minimal code changes! We'll make use of the principle of [_streaming_](https://huggingface.co/docs/datasets/stream), depicted graphically in Figure 1. Streaming does exactly this: the data is loaded progressively as we iterate over the dataset, meaning it is only loaded as and when we need it. If you're familiar with 🤗 Transformers and Datasets, the content of this notebook will be very familiar, with some small extensions to support streaming mode."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1c87f76e-47be-4a5d-bc52-7b1c2e9d4f5a",
   "metadata": {},
   "source": [
    "<figure>\n",
    "<img src=\"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/datasets/streaming.gif\" alt=\"Trulli\" style=\"width:100%\">\n",
    "<figcaption align = \"center\"><b>Figure 1:</b> Streaming mode. The dataset is divided into smaller subsets, with subsets loaded progressively as we iterate over the dataset. </figcaption>\n",
    "</figure>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "21b6316e-8a55-4549-a154-66d3da2ab74a",
   "metadata": {},
   "source": [
    "This notebook provides a guide to fine-tuning on the task of _speech recognition_, which involves learning a\n",
    "mapping from speech to text. Speech recognition is divided into two categories: English-only or multilingual (all other languages). \n",
    "This notebook applies to both categories, with pointers for changing between languages and datasets.\n",
    "\n",
    "As for our model, we'll fine-tune the Whisper model released in [September 2022](https://openai.com/blog/whisper/) by the authors \n",
    "Alec Radford et al. from OpenAI. Whisper is an encoder-decoder model pre-trained on 680k hours of labelled audio-transcription data. \n",
    "It achieves strong performance on many speech recognition and speech translation datasets without fine-tuning. With fine-tuning, \n",
    "we aim to improve upon these results further, with many SoTA results up for grabs! For a full explanation on the Whisper model, the \n",
    "reader is advised to read the blog post [Fine-Tune Whisper with 🤗 Transformers](https://huggingface.co/blog/fine-tune-whisper#introduction).\n",
    "\n",
    "The Whisper checkpoints come in five configurations of varying model sizes.\n",
    "The smallest four are trained on either English-only or multilingual data.\n",
    "The largest checkpoint is multilingual only. All nine of the pre-trained checkpoints \n",
    "are available on the [Hugging Face Hub](https://huggingface.co/models?search=openai/whisper). The \n",
    "checkpoints are summarised in the following table with links to the models on the Hub:\n",
    "\n",
    "| Size   | Layers | Width | Heads | Parameters | English-only                                         | Multilingual                                      |\n",
    "|--------|--------|-------|-------|------------|------------------------------------------------------|---------------------------------------------------|\n",
    "| tiny   | 4      | 384   | 6     | 39 M       | [✓](https://huggingface.co/openai/whisper-tiny.en)   | [✓](https://huggingface.co/openai/whisper-tiny.)  |\n",
    "| base   | 6      | 512   | 8     | 74 M       | [✓](https://huggingface.co/openai/whisper-base.en)   | [✓](https://huggingface.co/openai/whisper-base)   |\n",
    "| small  | 12     | 768   | 12    | 244 M      | [✓](https://huggingface.co/openai/whisper-small.en)  | [✓](https://huggingface.co/openai/whisper-small)  |\n",
    "| medium | 24     | 1024  | 16    | 769 M      | [✓](https://huggingface.co/openai/whisper-medium.en) | [✓](https://huggingface.co/openai/whisper-medium) |\n",
    "| large  | 32     | 1280  | 20    | 1550 M     | x                                                    | [✓](https://huggingface.co/openai/whisper-large)  |\n",
    "\n",
    "When fine-tuning on an English dataset for speech recognition, it is recommeneded to select one of the English-only checkpoints. For any other language, it is recommended to select a multilingual checkpoint.\n",
    "\n",
    "For demonstration purposes, we'll fine-tune the multilingual version of the \n",
    "[`\"small\"`](https://huggingface.co/openai/whisper-small) checkpoint with 244M params (~= 1GB). \n",
    "As for our data, we'll train and evaluate our system on 400 hours of multilingual speech recognition data\n",
    "taken from the [Common Voice 11](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0)\n",
    "dataset. We'll show how we can train a model on 400 hours of training data using the default disk space \n",
    "that comes with a standard GPU device or Google Colab."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b219c9dd-39b6-4a95-b2a1-3f547a1e7bc0",
   "metadata": {},
   "source": [
    "## Load Dataset with Streaming"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b17a4763-4381-4157-ae38-b04a8b5f1c43",
   "metadata": {},
   "source": [
    "This is where the magic happens! We'll first write a wrapper function around 🤗 Datasets `load_dataset` method. This function downloads the required splits using streaming mode by forcing `streaming=True` in the `load_dataset` method. Multiple splits can be combined (interleaved) by concatenating them with the \"+\" symbol when specifying the split name, e.g. `split=train+validation` will return a single split with the training and validation splits interleaved together. The function has the same arguments and key-word arguments as 🤗 Datasets `load_dataset` method, so we can use it in exactly the same way!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "065a8cf7-e54f-4ac3-900e-609c80714fca",
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import interleave_datasets, load_dataset\n",
    "\n",
    "def load_streaming_dataset(dataset_name, dataset_config_name, split, **kwargs):\n",
    "    if \"+\" in split:\n",
    "        # load multiple splits separated by the `+` symbol *with* streaming mode\n",
    "        dataset_splits = [load_dataset(dataset_name, dataset_config_name, split=split_name, streaming=True, **kwargs) for split_name in split.split(\"+\")]\n",
    "        # interleave multiple splits to form one dataset\n",
    "        interleaved_dataset = interleave_datasets(dataset_splits)\n",
    "        return interleaved_dataset\n",
    "    else:\n",
    "        # load a single split *with* streaming mode\n",
    "        dataset = load_dataset(dataset_name, dataset_config_name, split=split, streaming=True, **kwargs)\n",
    "        return dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "674429c5-0ab4-4adf-975b-621bb69eca38",
   "metadata": {},
   "source": [
    "We'll train our system on the Spanish split of [Common Voice 11](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0). We can see how much training data we have by viewing the [language page](https://commonvoice.mozilla.org/en/datasets) on the Common Voice website. The Spanish split has over 400 hours of labelled training data - that's enourmous! More than we could ever fit on a Google Colab or a standard workstation. But with streaming mode, we'll only download data as and when we need it, making training on this dataset possible!\n",
    "\n",
    "Since Spanish is relatively high-resource, we'll only use the `train` split for training and the `test` split for evaluation. If you're training on a low-resource language, such as the Hindi split of Common Voice 11, it's worth combining the `train` and `validation` splits to give a larger training set. You can achieve this by setting: `split=\"train+validation\"` for the training split.\n",
    "\n",
    "If you're using a gated dataset, like Common Voice 11, ensure you have accepted the terms of use on the Hugging Face Hub: [mozilla-foundation/common_voice_11_0](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0). Once you have accepted the terms, you will have full access to the dataset and be able to load the data locally."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a2787582-554f-44ce-9f38-4180a5ed6b44",
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import IterableDatasetDict\n",
    "\n",
    "raw_datasets = IterableDatasetDict()\n",
    "\n",
    "raw_datasets[\"train\"] = load_streaming_dataset(\"mozilla-foundation/common_voice_11_0\", \"es\", split=\"train\", use_auth_token=True)  # set split=\"train+validation\" for low-resource\n",
    "raw_datasets[\"test\"] = load_streaming_dataset(\"mozilla-foundation/common_voice_11_0\", \"es\", split=\"test\", use_auth_token=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2d63b2d2-f68a-4d74-b7f1-5127f6d16605",
   "metadata": {},
   "source": [
    "## Prepare Processor and Pre-Process Data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "601c3099-1026-439e-93e2-5635b3ba5a73",
   "metadata": {},
   "source": [
    "The ASR pipeline can be de-composed into three stages: \n",
    "1) A feature extractor which pre-processes the raw audio-inputs\n",
    "2) The model which performs the sequence-to-sequence mapping \n",
    "3) A tokenizer which post-processes the model outputs to text format\n",
    "\n",
    "In 🤗 Transformers, the Whisper model has an associated feature extractor and tokenizer, \n",
    "called [WhisperFeatureExtractor](https://huggingface.co/docs/transformers/main/model_doc/whisper#transformers.WhisperFeatureExtractor)\n",
    "and [WhisperTokenizer](https://huggingface.co/docs/transformers/main/model_doc/whisper#transformers.WhisperTokenizer) \n",
    "respectively. To make our lives simple, these two objects are wrapped under a single class, called the [WhisperProcessor](https://huggingface.co/docs/transformers/model_doc/whisper#transformers.WhisperProcessor). We can call the WhisperProcessor to perform \n",
    "both the audio pre-processing and the text token post-processing. In doing so, we only need to keep track of two objects during training: \n",
    "the `processor` and the `model`.\n",
    "\n",
    "If using a multilingual checkpoint, you should set the `\"language\"` to your target text language. You should also set the task to `\"transcribe\"` for speech recogntition and `\"translate\"` for speech translation. These arguments modify the behaviour of the tokenizer - they should be set correctly to ensure the target labels are encoded properly. These arguments should be omitted for English-only fine-tuning."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "77d9f0c5-8607-4642-a8ac-c3ab2e223ea6",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import WhisperProcessor\n",
    "\n",
    "processor = WhisperProcessor.from_pretrained(\"openai/whisper-small\", language=\"Spanish\", task=\"transcribe\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "381acd09-0b0f-4d04-9eb3-f028ac0e5f2c",
   "metadata": {},
   "source": [
    "### Pre-Process Data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bf10cd3e-924e-44fc-8790-46e413de7b3d",
   "metadata": {},
   "source": [
    "Let's have a look at the dataset features. Pay particular attention to the `\"audio\"` column - this details the sampling rate of our audio inputs:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ab5a13b4-9bd4-4aa0-aef2-b3de9b762988",
   "metadata": {},
   "outputs": [],
   "source": [
    "raw_datasets[\"train\"].features"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5a679f05-063d-41b3-9b58-4fc9c6ccf4fd",
   "metadata": {},
   "source": [
    "Since our input audio is sampled at 48kHz, we need to _downsample_ it to\n",
    "16kHz prior to passing it to the Whisper feature extractor, 16kHz being the sampling rate expected by the Whisper model. \n",
    "\n",
    "We'll set the audio inputs to the correct sampling rate using dataset's \n",
    "[`cast_column`](https://huggingface.co/docs/datasets/package_reference/main_classes.html?highlight=cast_column#datasets.DatasetDict.cast_column)\n",
    "method. This operation does not change the audio in-place, \n",
    "but rather signals to `datasets` to resample audio samples _on the fly_ the \n",
    "first time that they are loaded:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3ab6a724-3d1e-478b-a9e9-d2f85feb6c39",
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import Audio\n",
    "\n",
    "raw_datasets = raw_datasets.cast_column(\"audio\", Audio(sampling_rate=16000))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "161322c2-94f3-4d26-9e1d-d9d5202ca3cf",
   "metadata": {},
   "source": [
    "We'll define our pre-processing strategy. We advise that you **do not** lower-case the transcriptions or remove punctuation unless mixing different datasets. This will enable you to fine-tune Whisper models that can predict punctuation and casing. Later, you will see how we can evaluate the predictions without punctuation or casing, so that the models benefit from the WER improvement obtained by normalising the transcriptions while still predicting fully formatted transcriptions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d041650e-1c48-4439-87b3-5b6f4a514107",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers.models.whisper.english_normalizer import BasicTextNormalizer\n",
    "\n",
    "do_lower_case = False\n",
    "do_remove_punctuation = False\n",
    "\n",
    "normalizer = BasicTextNormalizer()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bfaa935b-a11d-497c-88c1-0c4d1bb3247b",
   "metadata": {},
   "source": [
    "Now we can write a function to prepare our data ready for the model:\n",
    "1. We load and resample the audio data by calling `batch[\"audio\"]`. As explained above, 🤗 Datasets performs any necessary resampling operations on the fly.\n",
    "2. We use the feature extractor to compute the log-Mel spectrogram input features from our 1-dimensional audio array.\n",
    "3. We perform any optional pre-processing (lower-case or remove punctuation).\n",
    "4. We encode the transcriptions to label ids through the use of the tokenizer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c085911c-a10a-41ef-8874-306e0503e9bb",
   "metadata": {},
   "outputs": [],
   "source": [
    "def prepare_dataset(batch):\n",
    "    # load and (possibly) resample audio data to 16kHz\n",
    "    audio = batch[\"audio\"]\n",
    "\n",
    "    # compute log-Mel input features from input audio array \n",
    "    batch[\"input_features\"] = processor.feature_extractor(audio[\"array\"], sampling_rate=audio[\"sampling_rate\"]).input_features[0]\n",
    "    # compute input length of audio sample in seconds\n",
    "    batch[\"input_length\"] = len(audio[\"array\"]) / audio[\"sampling_rate\"]\n",
    "    \n",
    "    # optional pre-processing steps\n",
    "    transcription = batch[\"sentence\"]\n",
    "    if do_lower_case:\n",
    "        transcription = transcription.lower()\n",
    "    if do_remove_punctuation:\n",
    "        transcription = normalizer(transcription).strip()\n",
    "    \n",
    "    # encode target text to label ids\n",
    "    batch[\"labels\"] = processor.tokenizer(transcription).input_ids\n",
    "    return batch"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "70b319fb-2439-4ef6-a70d-a47bf41c4a13",
   "metadata": {},
   "source": [
    "We can apply the data preparation function to all of our training examples using 🤗 Datasets' `.map` method. We'll remove all of the columns from the raw training data, leaving just the `input_features` and `labels` defined in the `prepare_dataset` function:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a37a7cdb-9013-427f-8de9-6a8d0e9dc684",
   "metadata": {},
   "outputs": [],
   "source": [
    "vectorized_datasets = raw_datasets.map(prepare_dataset, remove_columns=list(next(iter(raw_datasets.values())).features)).with_format(\"torch\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d59b37e-4950-47ec-9e3e-2cf2ec7fc750",
   "metadata": {},
   "source": [
    "We can now define how we shuffle the data in the train split. The size of the subset we load is set by the variable `buffer_size`. You can increase or decrease this depending on your memory constraints. In this example, the `buffer_size` is set to 500, meaning 500 samples are loaded before shuffling across the subset. The larger we set this value, the closer to True offline shuffling. The `seed` is set for reproducibility:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1b145699-acfc-4b1d-93a2-a2ad3d62674c",
   "metadata": {},
   "outputs": [],
   "source": [
    "vectorized_datasets[\"train\"] = vectorized_datasets[\"train\"].shuffle(\n",
    "    buffer_size=500,\n",
    "    seed=0,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "666b9ef0-7909-4e1e-a419-87604d233e29",
   "metadata": {},
   "source": [
    "Finally, we filter any training data with audio samples longer than 30s. These samples would otherwise be truncated by the Whisper feature-extractor which could affect the stability of training. We define a function that returns `True` for samples that are less than 30s, and `False` for those that are longer:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "01cb25ef-4bb0-4325-9461-f59198acadf6",
   "metadata": {},
   "outputs": [],
   "source": [
    "max_input_length = 30.0\n",
    "\n",
    "def is_audio_in_length_range(length):\n",
    "    return length < max_input_length"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "28e37ac3-b1c5-465b-8586-7cfd8d76b0f1",
   "metadata": {},
   "source": [
    "We apply our filter function to all samples of our training dataset through 🤗 Datasets' `.filter` method:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "333f7f6e-6053-4d3b-8924-c733c79b82ac",
   "metadata": {},
   "outputs": [],
   "source": [
    "vectorized_datasets[\"train\"] = vectorized_datasets[\"train\"].filter(\n",
    "    is_audio_in_length_range,\n",
    "    input_columns=[\"input_length\"],\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "263a5a58-0239-4a25-b0df-c625fc9c5810",
   "metadata": {},
   "source": [
    "## Training and Evaluation"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a693e768-c5a6-453f-89a1-b601dcf7daf7",
   "metadata": {},
   "source": [
    "Now that we've prepared our data, we're ready to dive into the training pipeline. \n",
    "The [🤗 Trainer](https://huggingface.co/transformers/master/main_classes/trainer.html?highlight=trainer)\n",
    "will do much of the heavy lifting for us. All we have to do is:\n",
    "\n",
    "- Define a data collator: the data collator takes our pre-processed data and prepares PyTorch tensors ready for the model.\n",
    "\n",
    "- Evaluation metrics: during evaluation, we want to evaluate the model using the [word error rate (WER)](https://huggingface.co/metrics/wer) metric. We need to define a `compute_metrics` function that handles this computation.\n",
    "\n",
    "- Load a pre-trained checkpoint: we need to load a pre-trained checkpoint and configure it correctly for training.\n",
    "\n",
    "- Define the training configuration: this will be used by the 🤗 Trainer to define the training schedule."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8d230e6d-624c-400a-bbf5-fa660881df25",
   "metadata": {},
   "source": [
    "### Define a Data Collator"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "04def221-0637-4a69-b242-d3f0c1d0ee78",
   "metadata": {},
   "source": [
    "The data collator for a sequence-to-sequence speech model is unique in the sense that it \n",
    "treats the `input_features` and `labels` independently: the  `input_features` must be \n",
    "handled by the feature extractor and the `labels` by the tokenizer.\n",
    "\n",
    "The `input_features` are already padded to 30s and converted to a log-Mel spectrogram \n",
    "of fixed dimension by action of the feature extractor, so all we have to do is convert the `input_features`\n",
    "to batched PyTorch tensors. We do this using the feature extractor's `.pad` method with `return_tensors=pt`.\n",
    "\n",
    "The `labels` on the other hand are un-padded. We first pad the sequences\n",
    "to the maximum length in the batch using the tokenizer's `.pad` method. The padding tokens \n",
    "are then replaced by `-100` so that these tokens are **not** taken into account when \n",
    "computing the loss. We then cut the BOS token from the start of the label sequence as we \n",
    "append it later during training.\n",
    "\n",
    "We can leverage the `WhisperProcessor` we defined earlier to perform both the \n",
    "feature extractor and the tokenizer operations:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8326221e-ec13-4731-bb4e-51e5fc1486c5",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "\n",
    "from dataclasses import dataclass\n",
    "from typing import Any, Dict, List, Union\n",
    "\n",
    "@dataclass\n",
    "class DataCollatorSpeechSeq2SeqWithPadding:\n",
    "    processor: Any\n",
    "\n",
    "    def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:\n",
    "        # split inputs and labels since they have to be of different lengths and need different padding methods\n",
    "        # first treat the audio inputs by simply returning torch tensors\n",
    "        input_features = [{\"input_features\": feature[\"input_features\"]} for feature in features]\n",
    "        batch = self.processor.feature_extractor.pad(input_features, return_tensors=\"pt\")\n",
    "\n",
    "        # get the tokenized label sequences\n",
    "        label_features = [{\"input_ids\": feature[\"labels\"]} for feature in features]\n",
    "        # pad the labels to max length\n",
    "        labels_batch = self.processor.tokenizer.pad(label_features, return_tensors=\"pt\")\n",
    "\n",
    "        # replace padding with -100 to ignore loss correctly\n",
    "        labels = labels_batch[\"input_ids\"].masked_fill(labels_batch.attention_mask.ne(1), -100)\n",
    "\n",
    "        # if bos token is appended in previous tokenization step,\n",
    "        # cut bos token here as it's append later anyways\n",
    "        if (labels[:, 0] == self.processor.tokenizer.bos_token_id).all().cpu().item():\n",
    "            labels = labels[:, 1:]\n",
    "\n",
    "        batch[\"labels\"] = labels\n",
    "\n",
    "        return batch"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3cae7dbf-8a50-456e-a3a8-7fd005390f86",
   "metadata": {},
   "source": [
    "Let's initialise the data collator we've just defined:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fc834702-c0d3-4a96-b101-7b87be32bf42",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor=processor)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d62bb2ab-750a-45e7-82e9-61d6f4805698",
   "metadata": {},
   "source": [
    "### Evaluation Metrics"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "66fee1a7-a44c-461e-b047-c3917221572e",
   "metadata": {},
   "source": [
    "We'll use the word error rate (WER) metric, the 'de-facto' metric for assessing \n",
    "ASR systems. For more information, refer to the WER [docs](https://huggingface.co/metrics/wer). We'll load the WER metric from 🤗 Evaluate:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b22b4011-f31f-4b57-b684-c52332f92890",
   "metadata": {},
   "outputs": [],
   "source": [
    "import evaluate\n",
    "\n",
    "metric = evaluate.load(\"wer\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "509f96d7-3f11-4f37-add9-f74a0c44f3fc",
   "metadata": {},
   "source": [
    "We then simply have to define a function that takes our model \n",
    "predictions and returns the WER metric. This function, called\n",
    "`compute_metrics`, first replaces `-100` with the `pad_token_id`\n",
    "in the `label_ids` (undoing the step we applied in the \n",
    "data collator to ignore padded tokens correctly in the loss).\n",
    "It then decodes the predicted and label ids to strings. Finally,\n",
    "it computes the WER between the predictions and reference labels. \n",
    "Here, we have the option of evaluating with the 'normalised' transcriptions \n",
    "and predictions. We recommend you set this to `True` to benefit from the WER \n",
    "improvement obtained by normalising the transcriptions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a11d1bfc-9e28-460f-a287-72d8f7bc1acb",
   "metadata": {},
   "outputs": [],
   "source": [
    "# evaluate with the 'normalised' WER\n",
    "do_normalize_eval = True\n",
    "\n",
    "def compute_metrics(pred):\n",
    "    pred_ids = pred.predictions\n",
    "    label_ids = pred.label_ids\n",
    "\n",
    "    # replace -100 with the pad_token_id\n",
    "    label_ids[label_ids == -100] = processor.tokenizer.pad_token_id\n",
    "\n",
    "    # we do not want to group tokens when computing the metrics\n",
    "    pred_str = processor.tokenizer.batch_decode(pred_ids, skip_special_tokens=True)\n",
    "    label_str = processor.tokenizer.batch_decode(label_ids, skip_special_tokens=True)\n",
    "\n",
    "    if do_normalize_eval:\n",
    "        pred_str = [normalizer(pred) for pred in pred_str]\n",
    "        label_str = [normalizer(label) for label in label_str]\n",
    "        # filtering step to only evaluate the samples that correspond to non-zero references:\n",
    "        pred_str = [pred_str[i] for i in range(len(pred_str)) if len(label_str[i]) > 0]\n",
    "        label_str = [label_str[i] for i in range(len(label_str)) if len(label_str[i]) > 0]\n",
    "\n",
    "    wer = 100 * metric.compute(predictions=pred_str, references=label_str)\n",
    "\n",
    "    return {\"wer\": wer}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "daf2a825-6d9f-4a23-b145-c37c0039075b",
   "metadata": {},
   "source": [
    "### Load a Pre-Trained Checkpoint"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "437a97fa-4864-476b-8abc-f28b8166cfa5",
   "metadata": {},
   "source": [
    "Now let's load the pre-trained Whisper `small` checkpoint. Again, this \n",
    "is trivial through use of 🤗 Transformers!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5a10cc4b-07ec-4ebd-ac1d-7c601023594f",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import WhisperForConditionalGeneration\n",
    "\n",
    "model = WhisperForConditionalGeneration.from_pretrained(\"openai/whisper-small\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a15ead5f-2277-4a39-937b-585c2497b2df",
   "metadata": {},
   "source": [
    "Override generation arguments - no tokens are forced as decoder outputs (see [`forced_decoder_ids`](https://huggingface.co/docs/transformers/main_classes/text_generation#transformers.generation_utils.GenerationMixin.generate.forced_decoder_ids)), no tokens are suppressed during generation (see [`suppress_tokens`](https://huggingface.co/docs/transformers/main_classes/text_generation#transformers.generation_utils.GenerationMixin.generate.suppress_tokens)). Set `use_cache` to False since we're using gradient checkpointing, and the two are incompatible:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "62038ba3-88ed-4fce-84db-338f50dcd04f",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.config.forced_decoder_ids = None\n",
    "model.config.suppress_tokens = []\n",
    "model.config.use_cache = False"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2178dea4-80ca-47b6-b6ea-ba1915c90c06",
   "metadata": {},
   "source": [
    "### Define the Training Configuration"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c21af1e9-0188-4134-ac82-defc7bdcc436",
   "metadata": {},
   "source": [
    "In the final step, we define all the parameters related to training. Here, you can set the `max_steps` to train for longer. For more detail on the training arguments, refer to the Seq2SeqTrainingArguments [docs](https://huggingface.co/docs/transformers/main_classes/trainer#transformers.Seq2SeqTrainingArguments)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0ae3e9af-97b7-4aa0-ae85-20b23b5bcb3a",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import Seq2SeqTrainingArguments\n",
    "\n",
    "training_args = Seq2SeqTrainingArguments(\n",
    "    output_dir=\"./\",\n",
    "    per_device_train_batch_size=64,\n",
    "    gradient_accumulation_steps=1,  # increase by 2x for every 2x decrease in batch size\n",
    "    learning_rate=1e-5,\n",
    "    warmup_steps=500,\n",
    "    max_steps=5000,\n",
    "    gradient_checkpointing=True,\n",
    "    fp16=True,\n",
    "    evaluation_strategy=\"steps\",\n",
    "    per_device_eval_batch_size=8,\n",
    "    predict_with_generate=True,\n",
    "    generation_max_length=225,\n",
    "    save_steps=1000,\n",
    "    eval_steps=1000,\n",
    "    logging_steps=25,\n",
    "    report_to=[\"tensorboard\"],\n",
    "    load_best_model_at_end=True,\n",
    "    metric_for_best_model=\"wer\",\n",
    "    greater_is_better=False,\n",
    "    push_to_hub=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3a944d8-3112-4552-82a0-be25988b3857",
   "metadata": {},
   "source": [
    "**Note**: if one does not want to upload the model checkpoints to the Hub, \n",
    "set `push_to_hub=False`."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "393c883e-3e50-492c-bd58-f51dbf15ee56",
   "metadata": {},
   "source": [
    "We then define a custom [Callback](https://huggingface.co/docs/transformers/main_classes/callback) that is called by the 🤗 Trainer on the end of each epoch. The Callback reinitialises and reshuffles the streaming dataset at the beginning of each new epoch - this gives different shuffling across our subsets for every epoch."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3ac16b62-b3c0-4c68-8f3d-9ecf471534b2",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import TrainerCallback\n",
    "from transformers.trainer_pt_utils import IterableDatasetShard\n",
    "from torch.utils.data import IterableDataset\n",
    "\n",
    "# trainer callback to reinitialise and reshuffle the streamable datasets at the beginning of each epoch\n",
    "class ShuffleCallback(TrainerCallback):\n",
    "    def on_epoch_begin(self, args, state, control, train_dataloader, **kwargs):\n",
    "        if isinstance(train_dataloader.dataset, IterableDatasetShard):\n",
    "            pass  # set_epoch() is handled by the Trainer\n",
    "        elif isinstance(train_dataloader.dataset, IterableDataset):\n",
    "            train_dataloader.dataset.set_epoch(train_dataloader.dataset._epoch + 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bac29114-d226-4f54-97cf-8718c9f94e1e",
   "metadata": {},
   "source": [
    "We can forward the training arguments to the 🤗 Trainer along with our model,\n",
    "dataset, data collator, `compute_metrics` function and custom callback:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d546d7fe-0543-479a-b708-2ebabec19493",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import Seq2SeqTrainer\n",
    "\n",
    "trainer = Seq2SeqTrainer(\n",
    "    args=training_args,\n",
    "    model=model,\n",
    "    train_dataset=vectorized_datasets[\"train\"],\n",
    "    eval_dataset=vectorized_datasets[\"test\"],\n",
    "    data_collator=data_collator,\n",
    "    compute_metrics=compute_metrics,\n",
    "    tokenizer=processor,\n",
    "    callbacks=[ShuffleCallback()],\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "67ab88c3-7091-4e51-8ad5-f5cacbe18449",
   "metadata": {},
   "source": [
    "We'll save the model and processor to the output directory before training:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a1ccb9ed-cbc8-4419-91c0-651e9424b672",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_pretrained(training_args.output_dir)\n",
    "processor.save_pretrained(training_args.output_dir)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7f404cf9-4345-468c-8196-4bd101d9bd51",
   "metadata": {},
   "source": [
    "### Training"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5e8b8d56-5a70-4f68-bd2e-f0752d0bd112",
   "metadata": {},
   "source": [
    "Training will take approximately 5-10 hours depending on your GPU. The peak GPU memory for the given training configuration is approximately 36GB. \n",
    "Depending on your GPU, it is possible that you will encounter a CUDA `\"out-of-memory\"` error when you launch training. \n",
    "In this case, you can reduce the `per_device_train_batch_size` incrementally by factors of 2 \n",
    "and employ [`gradient_accumulation_steps`](https://huggingface.co/docs/transformers/main_classes/trainer#transformers.Seq2SeqTrainingArguments.gradient_accumulation_steps)\n",
    "to compensate.\n",
    "\n",
    "To launch training, simply execute:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ee8b7b8e-1c9a-4d77-9137-1778a629e6de",
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer.train()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "747c6a6e",
   "metadata": {
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "(note that training may take some time to commence as we load the first training data samples with streaming mode)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "810ced54-7187-4a06-b2fe-ba6dcca94dc3",
   "metadata": {},
   "source": [
    "We can label our checkpoint with the `whisper-event` tag on push by setting the appropriate key-word arguments (kwargs):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6dd0e310-9b07-4133-ac14-2ed2d7524e22",
   "metadata": {},
   "outputs": [],
   "source": [
    "kwargs = {\n",
    "    \"dataset_tags\": \"mozilla-foundation/common_voice_11_0\",\n",
    "    \"dataset\": \"Common Voice 11.0\",  # a 'pretty' name for the training dataset\n",
    "    \"language\": \"es\",\n",
    "    \"model_name\": \"Whisper Small Es - Sanchit Gandhi\",  # a 'pretty' name for your model\n",
    "    \"finetuned_from\": \"openai/whisper-small\",\n",
    "    \"tasks\": \"automatic-speech-recognition\",\n",
    "    \"tags\": \"whisper-event\",\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "090d676a-f944-4297-a938-a40eda0b2b68",
   "metadata": {},
   "source": [
    "The training results can now be uploaded to the Hub. To do so, execute the `push_to_hub` command:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "95737cda-c5dd-4887-a4d0-dfcb0d61d977",
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer.push_to_hub(**kwargs)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}