Dongwei commited on
Commit
8a95a36
1 Parent(s): 4f4086f

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +790 -0
README.md ADDED
@@ -0,0 +1,790 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets:
3
+ - LIUM/tedlium
4
+ language:
5
+ - en
6
+ metrics:
7
+ - wer
8
+ library_name: espnet
9
+ pipeline_tag: automatic-speech-recognition
10
+ ---
11
+
12
+ <!-- Generated by scripts/utils/show_asr_result.sh -->
13
+ <!-- Generated by scripts/utils/show_asr_result.sh -->
14
+ # RESULTS
15
+ ## Environments
16
+ - date: `Mon Mar 27 04:02:03 EDT 2023`
17
+ - python version: `3.8.16 (default, Mar 2 2023, 03:21:46) [GCC 11.2.0]`
18
+ - espnet version: `espnet 202301`
19
+ - pytorch version: `pytorch 1.8.1`
20
+ - Git hash: `ff841366229d539eb74d23ac999cae7c0cc62cad`
21
+ - Commit date: `Mon Feb 20 12:23:15 2023 -0500`
22
+
23
+ ## exp/asr_train_raw_en_bpe500_sp
24
+ ### WER
25
+
26
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
27
+ |---|---|---|---|---|---|---|---|---|
28
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|14671|94.0|2.7|3.3|0.7|6.6|65.9|
29
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|27500|93.9|2.7|3.4|0.7|6.8|61.1|
30
+
31
+ ### CER
32
+
33
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
34
+ |---|---|---|---|---|---|---|---|---|
35
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|78259|96.6|0.6|2.8|0.6|4.0|65.9|
36
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|145066|96.6|0.6|2.8|0.6|4.1|61.1|
37
+
38
+ ### TER
39
+
40
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
41
+ |---|---|---|---|---|---|---|---|---|
42
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/dev|466|29364|95.5|1.9|2.7|0.5|5.1|65.9|
43
+ |decode_lm_lm_train_lm_en_bpe500_valid.loss.ave_asr_model_valid.acc.ave/test|1155|54206|95.5|1.7|2.7|0.6|5.1|61.1|
44
+
45
+
46
+ ## ASR config
47
+
48
+ <details><summary>expand</summary>
49
+
50
+ ```
51
+ config: conf/train.yaml
52
+ print_config: false
53
+ log_level: INFO
54
+ dry_run: false
55
+ iterator_type: sequence
56
+ output_dir: exp/asr_train_raw_en_bpe500_sp
57
+ ngpu: 1
58
+ seed: 2022
59
+ num_workers: 6
60
+ num_att_plot: 3
61
+ dist_backend: nccl
62
+ dist_init_method: env://
63
+ dist_world_size: 4
64
+ dist_rank: 0
65
+ local_rank: 0
66
+ dist_master_addr: localhost
67
+ dist_master_port: 46711
68
+ dist_launcher: null
69
+ multiprocessing_distributed: true
70
+ unused_parameters: false
71
+ sharded_ddp: false
72
+ cudnn_enabled: true
73
+ cudnn_benchmark: false
74
+ cudnn_deterministic: true
75
+ collect_stats: false
76
+ write_collected_feats: false
77
+ max_epoch: 50
78
+ patience: null
79
+ val_scheduler_criterion:
80
+ - valid
81
+ - loss
82
+ early_stopping_criterion:
83
+ - valid
84
+ - loss
85
+ - min
86
+ best_model_criterion:
87
+ - - valid
88
+ - acc
89
+ - max
90
+ keep_nbest_models: 10
91
+ nbest_averaging_interval: 0
92
+ grad_clip: 5.0
93
+ grad_clip_type: 2.0
94
+ grad_noise: false
95
+ accum_grad: 1
96
+ no_forward_run: false
97
+ resume: true
98
+ train_dtype: float32
99
+ use_amp: true
100
+ log_interval: null
101
+ use_matplotlib: true
102
+ use_tensorboard: true
103
+ create_graph_in_tensorboard: false
104
+ use_wandb: false
105
+ wandb_project: null
106
+ wandb_id: null
107
+ wandb_entity: null
108
+ wandb_name: null
109
+ wandb_model_log_interval: -1
110
+ detect_anomaly: false
111
+ pretrain_path: null
112
+ init_param: []
113
+ ignore_init_mismatch: false
114
+ freeze_param: []
115
+ num_iters_per_epoch: null
116
+ batch_size: 20
117
+ valid_batch_size: null
118
+ batch_bins: 50000000
119
+ valid_batch_bins: null
120
+ train_shape_file:
121
+ - exp/asr_stats_raw_en_bpe500_sp/train/speech_shape
122
+ - exp/asr_stats_raw_en_bpe500_sp/train/text_shape.bpe
123
+ valid_shape_file:
124
+ - exp/asr_stats_raw_en_bpe500_sp/valid/speech_shape
125
+ - exp/asr_stats_raw_en_bpe500_sp/valid/text_shape.bpe
126
+ batch_type: numel
127
+ valid_batch_type: null
128
+ fold_length:
129
+ - 80000
130
+ - 150
131
+ sort_in_batch: descending
132
+ sort_batch: descending
133
+ multiple_iterator: false
134
+ chunk_length: 500
135
+ chunk_shift_ratio: 0.5
136
+ num_cache_chunks: 1024
137
+ train_data_path_and_name_and_type:
138
+ - - dump/raw/train_sp/wav.scp
139
+ - speech
140
+ - kaldi_ark
141
+ - - dump/raw/train_sp/text
142
+ - text
143
+ - text
144
+ valid_data_path_and_name_and_type:
145
+ - - dump/raw/dev/wav.scp
146
+ - speech
147
+ - kaldi_ark
148
+ - - dump/raw/dev/text
149
+ - text
150
+ - text
151
+ allow_variable_data_keys: false
152
+ max_cache_size: 0.0
153
+ max_cache_fd: 32
154
+ valid_max_cache_size: null
155
+ exclude_weight_decay: false
156
+ exclude_weight_decay_conf: {}
157
+ optim: adam
158
+ optim_conf:
159
+ lr: 0.002
160
+ weight_decay: 1.0e-06
161
+ scheduler: warmuplr
162
+ scheduler_conf:
163
+ warmup_steps: 15000
164
+ token_list:
165
+ - <blank>
166
+ - <unk>
167
+ - '[unk]'
168
+ - ▁
169
+ - s
170
+ - ▁the
171
+ - t
172
+ - ▁and
173
+ - e
174
+ - ▁a
175
+ - ▁to
176
+ - d
177
+ - ▁of
178
+ - ''''
179
+ - n
180
+ - ing
181
+ - ▁in
182
+ - ▁that
183
+ - re
184
+ - ▁i
185
+ - c
186
+ - o
187
+ - u
188
+ - ▁we
189
+ - y
190
+ - a
191
+ - ed
192
+ - ▁it
193
+ - ▁you
194
+ - i
195
+ - m
196
+ - ▁is
197
+ - er
198
+ - p
199
+ - g
200
+ - w
201
+ - al
202
+ - ▁this
203
+ - ▁so
204
+ - f
205
+ - le
206
+ - b
207
+ - ar
208
+ - ▁f
209
+ - k
210
+ - ▁c
211
+ - r
212
+ - in
213
+ - or
214
+ - ▁for
215
+ - ▁be
216
+ - ve
217
+ - ▁was
218
+ - te
219
+ - th
220
+ - ▁do
221
+ - es
222
+ - ly
223
+ - ▁they
224
+ - ro
225
+ - ▁are
226
+ - ▁with
227
+ - ▁have
228
+ - an
229
+ - v
230
+ - ch
231
+ - ▁on
232
+ - se
233
+ - lo
234
+ - ▁but
235
+ - en
236
+ - ri
237
+ - li
238
+ - ▁what
239
+ - it
240
+ - ic
241
+ - ▁can
242
+ - l
243
+ - ur
244
+ - ce
245
+ - ent
246
+ - ▁me
247
+ - ▁b
248
+ - ▁ma
249
+ - ▁he
250
+ - ra
251
+ - ▁de
252
+ - ll
253
+ - at
254
+ - ▁about
255
+ - ▁one
256
+ - ▁not
257
+ - ne
258
+ - ▁all
259
+ - ▁my
260
+ - ter
261
+ - el
262
+ - il
263
+ - ▁there
264
+ - 'on'
265
+ - ad
266
+ - ▁mo
267
+ - ol
268
+ - ation
269
+ - nd
270
+ - ▁like
271
+ - ▁people
272
+ - po
273
+ - ▁at
274
+ - ▁us
275
+ - us
276
+ - ▁g
277
+ - ci
278
+ - ▁our
279
+ - h
280
+ - pe
281
+ - ▁as
282
+ - ▁from
283
+ - vi
284
+ - ▁if
285
+ - as
286
+ - ▁ex
287
+ - ▁con
288
+ - ▁an
289
+ - ver
290
+ - ▁out
291
+ - ▁just
292
+ - un
293
+ - ▁see
294
+ - la
295
+ - ▁di
296
+ - ▁when
297
+ - ▁now
298
+ - ▁p
299
+ - ha
300
+ - ▁who
301
+ - ck
302
+ - ▁these
303
+ - ▁because
304
+ - ▁or
305
+ - ▁know
306
+ - ion
307
+ - ir
308
+ - ▁co
309
+ - ▁up
310
+ - ▁pa
311
+ - ment
312
+ - ▁think
313
+ - ge
314
+ - ▁how
315
+ - ide
316
+ - ▁by
317
+ - ul
318
+ - ity
319
+ - ▁go
320
+ - ▁get
321
+ - ▁ho
322
+ - ive
323
+ - ▁very
324
+ - ate
325
+ - ng
326
+ - ▁no
327
+ - ▁had
328
+ - ac
329
+ - ▁bo
330
+ - ry
331
+ - ▁more
332
+ - ▁them
333
+ - ▁some
334
+ - mi
335
+ - ▁time
336
+ - ▁your
337
+ - me
338
+ - ▁going
339
+ - op
340
+ - am
341
+ - per
342
+ - et
343
+ - ▁would
344
+ - ru
345
+ - ure
346
+ - ti
347
+ - ist
348
+ - ▁their
349
+ - x
350
+ - ▁were
351
+ - ▁look
352
+ - ▁pro
353
+ - ▁which
354
+ - ▁work
355
+ - tion
356
+ - est
357
+ - ty
358
+ - im
359
+ - z
360
+ - ta
361
+ - ▁want
362
+ - ▁two
363
+ - age
364
+ - ▁really
365
+ - om
366
+ - ma
367
+ - ers
368
+ - ting
369
+ - ▁world
370
+ - co
371
+ - ▁way
372
+ - ▁don
373
+ - wa
374
+ - hi
375
+ - tra
376
+ - ▁la
377
+ - ▁here
378
+ - able
379
+ - lu
380
+ - ▁other
381
+ - mo
382
+ - ies
383
+ - ▁has
384
+ - ▁could
385
+ - j
386
+ - ▁make
387
+ - ally
388
+ - ▁sta
389
+ - ten
390
+ - ▁will
391
+ - ▁un
392
+ - ig
393
+ - ▁where
394
+ - ▁into
395
+ - ke
396
+ - ▁than
397
+ - ▁comp
398
+ - ▁actually
399
+ - tic
400
+ - sh
401
+ - ▁did
402
+ - tor
403
+ - fa
404
+ - ical
405
+ - ▁she
406
+ - ▁years
407
+ - ▁say
408
+ - one
409
+ - ted
410
+ - ▁things
411
+ - ph
412
+ - ▁new
413
+ - ▁pre
414
+ - ▁any
415
+ - ▁thousand
416
+ - ▁been
417
+ - ▁inter
418
+ - ▁his
419
+ - ▁com
420
+ - ▁need
421
+ - nce
422
+ - ▁right
423
+ - ▁take
424
+ - ▁even
425
+ - ▁over
426
+ - ▁start
427
+ - ▁hundred
428
+ - min
429
+ - ▁sp
430
+ - ▁those
431
+ - ▁car
432
+ - ▁then
433
+ - mp
434
+ - ap
435
+ - ▁first
436
+ - les
437
+ - ize
438
+ - ▁every
439
+ - ba
440
+ - ▁something
441
+ - ▁well
442
+ - ard
443
+ - ▁str
444
+ - ▁back
445
+ - und
446
+ - ia
447
+ - pl
448
+ - ki
449
+ - ho
450
+ - ▁call
451
+ - ▁most
452
+ - ▁also
453
+ - bi
454
+ - ▁thing
455
+ - ▁life
456
+ - um
457
+ - ▁said
458
+ - ▁kind
459
+ - ▁lot
460
+ - ▁much
461
+ - va
462
+ - ▁ra
463
+ - ▁little
464
+ - ▁dr
465
+ - ▁got
466
+ - ▁come
467
+ - ful
468
+ - ▁talk
469
+ - ▁part
470
+ - ▁day
471
+ - ant
472
+ - ction
473
+ - ▁happen
474
+ - ▁only
475
+ - ▁many
476
+ - ▁wo
477
+ - pri
478
+ - ▁her
479
+ - ▁br
480
+ - qui
481
+ - ▁mean
482
+ - ▁three
483
+ - iv
484
+ - ▁different
485
+ - ugh
486
+ - ain
487
+ - ▁human
488
+ - ance
489
+ - ▁change
490
+ - ▁let
491
+ - ▁real
492
+ - ▁show
493
+ - ▁good
494
+ - ▁around
495
+ - ▁through
496
+ - ▁jo
497
+ - bu
498
+ - ▁down
499
+ - ight
500
+ - ga
501
+ - ▁why
502
+ - ▁live
503
+ - ff
504
+ - ▁tell
505
+ - ▁put
506
+ - ▁idea
507
+ - port
508
+ - ▁same
509
+ - ▁give
510
+ - ated
511
+ - ish
512
+ - ible
513
+ - ▁though
514
+ - ious
515
+ - ▁problem
516
+ - ▁five
517
+ - par
518
+ - ▁fact
519
+ - ▁cha
520
+ - ition
521
+ - ▁year
522
+ - ▁big
523
+ - ▁plan
524
+ - ▁great
525
+ - ▁find
526
+ - ▁four
527
+ - ▁app
528
+ - ▁after
529
+ - ▁system
530
+ - ▁place
531
+ - ▁em
532
+ - ▁build
533
+ - ▁percent
534
+ - ▁again
535
+ - ▁point
536
+ - ▁learn
537
+ - ▁own
538
+ - ▁long
539
+ - ▁made
540
+ - ▁today
541
+ - ▁nine
542
+ - ities
543
+ - ▁gene
544
+ - ▁six
545
+ - ▁question
546
+ - light
547
+ - ▁should
548
+ - ▁came
549
+ - ▁feel
550
+ - ▁turn
551
+ - ▁person
552
+ - ▁end
553
+ - ▁hu
554
+ - ▁design
555
+ - ▁help
556
+ - ▁brain
557
+ - ▁last
558
+ - ▁create
559
+ - ▁important
560
+ - ▁before
561
+ - ▁high
562
+ - ▁never
563
+ - ▁trans
564
+ - ▁another
565
+ - ▁him
566
+ - ▁eight
567
+ - ▁might
568
+ - ▁understand
569
+ - ▁power
570
+ - ▁better
571
+ - q
572
+ - ▁found
573
+ - ▁play
574
+ - ▁twenty
575
+ - ▁still
576
+ - ▁school
577
+ - ▁each
578
+ - ▁seven
579
+ - ▁together
580
+ - ▁few
581
+ - ▁hand
582
+ - ▁example
583
+ - que
584
+ - ▁next
585
+ - ▁million
586
+ - ▁story
587
+ - ▁women
588
+ - ▁under
589
+ - ▁number
590
+ - ▁course
591
+ - ▁water
592
+ - ▁ago
593
+ - ▁grow
594
+ - ▁between
595
+ - ▁develop
596
+ - ▁america
597
+ - ▁sort
598
+ - ▁technology
599
+ - ▁believe
600
+ - ▁second
601
+ - ▁small
602
+ - ▁maybe
603
+ - ▁become
604
+ - press
605
+ - ▁health
606
+ - ▁space
607
+ - ▁word
608
+ - ▁hard
609
+ - ▁children
610
+ - ▁organ
611
+ - ▁always
612
+ - ▁country
613
+ - ▁reason
614
+ - ▁experience
615
+ - ▁large
616
+ - ▁everything
617
+ - ▁friend
618
+ - ▁project
619
+ - ▁computer
620
+ - ▁fifty
621
+ - ▁money
622
+ - ▁information
623
+ - graph
624
+ - ▁walk
625
+ - ization
626
+ - ▁africa
627
+ - ▁picture
628
+ - ▁process
629
+ - ▁teach
630
+ - ▁enough
631
+ - ▁elect
632
+ - ▁thirty
633
+ - '0'
634
+ - '1'
635
+ - '2'
636
+ - '9'
637
+ - '3'
638
+ - '5'
639
+ - '8'
640
+ - '4'
641
+ - '7'
642
+ - '6'
643
+ - '&'
644
+ - +
645
+ - '#'
646
+ - '@'
647
+ - '*'
648
+ - \
649
+ - ^
650
+ - R
651
+ - _
652
+ - '-'
653
+ - '%'
654
+ - '='
655
+ - $
656
+ - M
657
+ - ā
658
+ - ']'
659
+ - E
660
+ - U
661
+ - A
662
+ - G
663
+ - '['
664
+ - <sos/eos>
665
+ init: null
666
+ input_size: null
667
+ ctc_conf:
668
+ dropout_rate: 0.0
669
+ ctc_type: builtin
670
+ reduce: true
671
+ ignore_nan_grad: null
672
+ zero_infinity: true
673
+ joint_net_conf: null
674
+ use_preprocessor: true
675
+ token_type: bpe
676
+ bpemodel: data/en_token_list/bpe_unigram500/bpe.model
677
+ non_linguistic_symbols: null
678
+ cleaner: null
679
+ g2p: null
680
+ speech_volume_normalize: null
681
+ rir_scp: null
682
+ rir_apply_prob: 1.0
683
+ noise_scp: null
684
+ noise_apply_prob: 1.0
685
+ noise_db_range: '13_15'
686
+ short_noise_thres: 0.5
687
+ aux_ctc_tasks: []
688
+ frontend: default
689
+ frontend_conf:
690
+ n_fft: 512
691
+ win_length: 400
692
+ hop_length: 160
693
+ fs: 16k
694
+ specaug: specaug
695
+ specaug_conf:
696
+ apply_time_warp: true
697
+ time_warp_window: 5
698
+ time_warp_mode: bicubic
699
+ apply_freq_mask: true
700
+ freq_mask_width_range:
701
+ - 0
702
+ - 27
703
+ num_freq_mask: 2
704
+ apply_time_mask: true
705
+ time_mask_width_ratio_range:
706
+ - 0.0
707
+ - 0.05
708
+ num_time_mask: 5
709
+ normalize: global_mvn
710
+ normalize_conf:
711
+ stats_file: exp/asr_stats_raw_en_bpe500_sp/train/feats_stats.npz
712
+ model: espnet
713
+ model_conf:
714
+ ctc_weight: 0.3
715
+ lsm_weight: 0.1
716
+ length_normalized_loss: false
717
+ preencoder: null
718
+ preencoder_conf: {}
719
+ encoder: conformer
720
+ encoder_conf:
721
+ output_size: 256
722
+ attention_heads: 4
723
+ linear_units: 1024
724
+ num_blocks: 12
725
+ dropout_rate: 0.1
726
+ positional_dropout_rate: 0.1
727
+ attention_dropout_rate: 0.1
728
+ input_layer: conv2d
729
+ normalize_before: true
730
+ macaron_style: true
731
+ rel_pos_type: latest
732
+ pos_enc_layer_type: rel_pos
733
+ selfattention_layer_type: rel_selfattn
734
+ activation_type: swish
735
+ use_cnn_module: true
736
+ cnn_module_kernel: 31
737
+ postencoder: null
738
+ postencoder_conf: {}
739
+ decoder: transformer
740
+ decoder_conf:
741
+ attention_heads: 4
742
+ linear_units: 2048
743
+ num_blocks: 6
744
+ dropout_rate: 0.1
745
+ positional_dropout_rate: 0.1
746
+ self_attention_dropout_rate: 0.1
747
+ src_attention_dropout_rate: 0.1
748
+ preprocessor: default
749
+ preprocessor_conf: {}
750
+ required:
751
+ - output_dir
752
+ - token_list
753
+ version: '202301'
754
+ distributed: true
755
+ ```
756
+
757
+ </details>
758
+
759
+
760
+
761
+ ### Citing ESPnet
762
+
763
+ ```BibTex
764
+ @inproceedings{watanabe2018espnet,
765
+ author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
766
+ title={{ESPnet}: End-to-End Speech Processing Toolkit},
767
+ year={2018},
768
+ booktitle={Proceedings of Interspeech},
769
+ pages={2207--2211},
770
+ doi={10.21437/Interspeech.2018-1456},
771
+ url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
772
+ }
773
+
774
+
775
+
776
+
777
+ ```
778
+
779
+ or arXiv:
780
+
781
+ ```bibtex
782
+ @misc{watanabe2018espnet,
783
+ title={ESPnet: End-to-End Speech Processing Toolkit},
784
+ author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
785
+ year={2018},
786
+ eprint={1804.00015},
787
+ archivePrefix={arXiv},
788
+ primaryClass={cs.CL}
789
+ }
790
+ ```