“siddhu001” commited on
Commit
695b416
1 Parent(s): c6ad5ed

Update model

Browse files
README.md ADDED
@@ -0,0 +1,1359 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - espnet
4
+ - audio
5
+ - automatic-speech-recognition
6
+ language: en
7
+ datasets:
8
+ - slue-voxceleb
9
+ license: cc-by-4.0
10
+ ---
11
+
12
+ ## ESPnet2 ASR model
13
+
14
+ ### `espnet/sluevoxceleb_owsm_lightweight_asr`
15
+
16
+ This model was trained by “siddhu001” using slue-voxceleb recipe in [espnet](https://github.com/espnet/espnet/).
17
+
18
+ ### Demo: How to use in ESPnet2
19
+
20
+ Follow the [ESPnet installation instructions](https://espnet.github.io/espnet/installation.html)
21
+ if you haven't done that already.
22
+
23
+ ```bash
24
+ cd espnet
25
+ git checkout e23ef85f0b3116ad5c60d0833f186da0deec0734
26
+ pip install -e .
27
+ cd egs2/slue-voxceleb/slu1_asr
28
+ ./run.sh --skip_data_prep false --skip_train true --download_model espnet/sluevoxceleb_owsm_lightweight_asr
29
+ ```
30
+
31
+ <!-- Generated by scripts/utils/show_asr_result.sh -->
32
+ # RESULTS
33
+ ## Environments
34
+ - date: `Mon Feb 5 15:08:12 CST 2024`
35
+ - python version: `3.9.13 (main, Aug 25 2022, 23:26:10) [GCC 11.2.0]`
36
+ - espnet version: `espnet 202310`
37
+ - pytorch version: `pytorch 2.1.0+cu121`
38
+ - Git hash: `21d2105784e4da98397bf487b2550d4c6e16d40d`
39
+ - Commit date: `Wed Jan 31 13:40:37 2024 -0600`
40
+
41
+ ## exp/slu_train_asr_owsm_superb_raw_en_word_sp
42
+ ### WER
43
+
44
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
45
+ |---|---|---|---|---|---|---|---|---|
46
+ |decode_asr_ctc_slu_model_valid.cer_ctc.ave/test|3426|135368|87.2|7.7|5.1|3.5|16.3|93.9|
47
+
48
+ ### CER
49
+
50
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
51
+ |---|---|---|---|---|---|---|---|---|
52
+ |decode_asr_ctc_slu_model_valid.cer_ctc.ave/test|3426|591261|93.9|1.8|4.4|3.3|9.5|93.9|
53
+
54
+ ### TER
55
+
56
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
57
+ |---|---|---|---|---|---|---|---|---|
58
+ ## exp/slu_train_asr_owsm_superb_raw_en_word_sp/decode_asr_ctc_slu_model_valid.cer_ctc.ave
59
+ ### WER
60
+
61
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
62
+ |---|---|---|---|---|---|---|---|---|
63
+ |org/devel|1437|56031|89.2|6.3|4.4|3.1|13.9|92.6|
64
+
65
+ ### CER
66
+
67
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
68
+ |---|---|---|---|---|---|---|---|---|
69
+ |org/devel|1437|241556|95.0|1.3|3.6|2.9|7.9|92.6|
70
+
71
+ ### TER
72
+
73
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
74
+ |---|---|---|---|---|---|---|---|---|
75
+
76
+ ## ASR config
77
+
78
+ <details><summary>expand</summary>
79
+
80
+ ```
81
+ config: conf/tuning/train_asr_owsm_superb.yaml
82
+ print_config: false
83
+ log_level: INFO
84
+ drop_last_iter: false
85
+ dry_run: false
86
+ iterator_type: sequence
87
+ valid_iterator_type: null
88
+ output_dir: exp/slu_train_asr_owsm_superb_raw_en_word_sp
89
+ ngpu: 1
90
+ seed: 2022
91
+ num_workers: 2
92
+ num_att_plot: 3
93
+ dist_backend: nccl
94
+ dist_init_method: env://
95
+ dist_world_size: 4
96
+ dist_rank: 0
97
+ local_rank: 0
98
+ dist_master_addr: localhost
99
+ dist_master_port: 48677
100
+ dist_launcher: null
101
+ multiprocessing_distributed: true
102
+ unused_parameters: false
103
+ sharded_ddp: false
104
+ cudnn_enabled: true
105
+ cudnn_benchmark: false
106
+ cudnn_deterministic: true
107
+ collect_stats: false
108
+ write_collected_feats: false
109
+ max_epoch: 70
110
+ patience: null
111
+ val_scheduler_criterion:
112
+ - valid
113
+ - loss
114
+ early_stopping_criterion:
115
+ - valid
116
+ - loss
117
+ - min
118
+ best_model_criterion:
119
+ - - valid
120
+ - cer_ctc
121
+ - min
122
+ - - valid
123
+ - loss
124
+ - min
125
+ keep_nbest_models: 10
126
+ nbest_averaging_interval: 0
127
+ grad_clip: 5.0
128
+ grad_clip_type: 2.0
129
+ grad_noise: false
130
+ accum_grad: 2
131
+ no_forward_run: false
132
+ resume: true
133
+ train_dtype: float32
134
+ use_amp: false
135
+ log_interval: null
136
+ use_matplotlib: true
137
+ use_tensorboard: true
138
+ create_graph_in_tensorboard: false
139
+ use_wandb: false
140
+ wandb_project: null
141
+ wandb_id: null
142
+ wandb_entity: null
143
+ wandb_name: null
144
+ wandb_model_log_interval: -1
145
+ detect_anomaly: false
146
+ use_lora: false
147
+ save_lora_only: true
148
+ lora_conf: {}
149
+ pretrain_path: null
150
+ init_param:
151
+ - /scratch/bbjs/arora1/new_download_espnet_egs2/harpervalley/slu1_superb_onlyda/owsm_v3.1_ebf/exp/s2t_train_s2t_ebf_conv2d_size1024_e18_d18_piecewise_lr2e-4_warmup60k_flashattn_raw_bpe50000/valid.total_count.ave_5best.till45epoch.pth:encoder:encoder
152
+ ignore_init_mismatch: false
153
+ freeze_param:
154
+ - encoder
155
+ num_iters_per_epoch: null
156
+ batch_size: 20
157
+ valid_batch_size: null
158
+ batch_bins: 6000000
159
+ valid_batch_bins: null
160
+ train_shape_file:
161
+ - exp/slu_stats_raw_en_word_sp/train/speech_shape
162
+ - exp/slu_stats_raw_en_word_sp/train/text_shape.word
163
+ valid_shape_file:
164
+ - exp/slu_stats_raw_en_word_sp/valid/speech_shape
165
+ - exp/slu_stats_raw_en_word_sp/valid/text_shape.word
166
+ batch_type: numel
167
+ valid_batch_type: null
168
+ fold_length:
169
+ - 80000
170
+ - 150
171
+ sort_in_batch: descending
172
+ shuffle_within_batch: false
173
+ sort_batch: descending
174
+ multiple_iterator: false
175
+ chunk_length: 500
176
+ chunk_shift_ratio: 0.5
177
+ num_cache_chunks: 1024
178
+ chunk_excluded_key_prefixes: []
179
+ chunk_default_fs: null
180
+ train_data_path_and_name_and_type:
181
+ - - dump/raw/train_sp/wav.scp
182
+ - speech
183
+ - sound
184
+ - - dump/raw/train_sp/text
185
+ - text
186
+ - text
187
+ valid_data_path_and_name_and_type:
188
+ - - dump/raw/devel/wav.scp
189
+ - speech
190
+ - sound
191
+ - - dump/raw/devel/text
192
+ - text
193
+ - text
194
+ allow_variable_data_keys: false
195
+ max_cache_size: 0.0
196
+ max_cache_fd: 32
197
+ allow_multi_rates: false
198
+ valid_max_cache_size: null
199
+ exclude_weight_decay: false
200
+ exclude_weight_decay_conf: {}
201
+ optim: adam
202
+ optim_conf:
203
+ lr: 0.01
204
+ weight_decay: 1.0e-06
205
+ scheduler: warmuplr
206
+ scheduler_conf:
207
+ warmup_steps: 5000
208
+ token_list:
209
+ - <blank>
210
+ - <unk>
211
+ - ▁i
212
+ - ▁and
213
+ - ''''
214
+ - s
215
+ - ▁the
216
+ - ▁a
217
+ - ▁it
218
+ - ▁to
219
+ - ▁you
220
+ - ▁that
221
+ - ▁of
222
+ - ▁in
223
+ - ▁was
224
+ - ▁uh
225
+ - ▁know
226
+ - t
227
+ - ▁so
228
+ - ▁we
229
+ - ▁he
230
+ - ing
231
+ - m
232
+ - ▁um
233
+ - ▁like
234
+ - ed
235
+ - ▁is
236
+ - ▁but
237
+ - ▁just
238
+ - ▁they
239
+ - re
240
+ - y
241
+ - ▁this
242
+ - ▁for
243
+ - ▁be
244
+ - ▁my
245
+ - er
246
+ - ▁with
247
+ - ▁on
248
+ - ▁think
249
+ - ▁have
250
+ - ▁p
251
+ - ▁she
252
+ - ▁me
253
+ - e
254
+ - ▁really
255
+ - ▁there
256
+ - ▁what
257
+ - al
258
+ - ▁m
259
+ - ▁do
260
+ - ▁all
261
+ - a
262
+ - ve
263
+ - ▁as
264
+ - c
265
+ - n
266
+ - ▁about
267
+ - ▁not
268
+ - i
269
+ - ▁at
270
+ - l
271
+ - ▁t
272
+ - ▁had
273
+ - ▁when
274
+ - ▁c
275
+ - g
276
+ - in
277
+ - ▁b
278
+ - d
279
+ - le
280
+ - en
281
+ - ▁out
282
+ - u
283
+ - ly
284
+ - ▁an
285
+ - or
286
+ - ▁people
287
+ - ar
288
+ - ll
289
+ - o
290
+ - ▁are
291
+ - ▁very
292
+ - ▁because
293
+ - es
294
+ - ▁can
295
+ - ▁don
296
+ - ▁s
297
+ - ▁or
298
+ - ▁up
299
+ - it
300
+ - b
301
+ - ▁e
302
+ - ▁one
303
+ - an
304
+ - st
305
+ - ▁if
306
+ - ▁f
307
+ - ▁were
308
+ - p
309
+ - ▁mean
310
+ - ▁d
311
+ - ▁who
312
+ - ▁then
313
+ - ic
314
+ - 'on'
315
+ - ▁no
316
+ - ▁go
317
+ - ▁her
318
+ - ▁g
319
+ - ▁st
320
+ - ▁kind
321
+ - ri
322
+ - ▁would
323
+ - ▁get
324
+ - at
325
+ - r
326
+ - ▁time
327
+ - v
328
+ - ent
329
+ - ▁re
330
+ - h
331
+ - ▁from
332
+ - ▁l
333
+ - ▁said
334
+ - ▁w
335
+ - ▁him
336
+ - ▁how
337
+ - ▁well
338
+ - ▁h
339
+ - ▁gonna
340
+ - ▁lot
341
+ - ▁see
342
+ - w
343
+ - ▁his
344
+ - ce
345
+ - ion
346
+ - ▁been
347
+ - f
348
+ - ▁great
349
+ - ▁yeah
350
+ - ▁love
351
+ - ▁which
352
+ - ▁got
353
+ - k
354
+ - ▁them
355
+ - ▁way
356
+ - ▁n
357
+ - id
358
+ - ▁show
359
+ - ▁some
360
+ - ▁your
361
+ - ▁did
362
+ - ▁sort
363
+ - et
364
+ - ▁has
365
+ - ▁things
366
+ - ▁back
367
+ - ▁where
368
+ - ▁something
369
+ - ir
370
+ - ▁thing
371
+ - ad
372
+ - ▁su
373
+ - il
374
+ - as
375
+ - ▁j
376
+ - ▁more
377
+ - ▁co
378
+ - se
379
+ - ▁say
380
+ - nd
381
+ - ▁much
382
+ - ▁come
383
+ - ▁always
384
+ - ine
385
+ - ▁r
386
+ - ation
387
+ - ▁other
388
+ - th
389
+ - ur
390
+ - ▁se
391
+ - ▁now
392
+ - ate
393
+ - ▁doing
394
+ - ▁work
395
+ - ow
396
+ - ▁could
397
+ - ally
398
+ - ▁these
399
+ - ▁good
400
+ - ▁any
401
+ - ▁cause
402
+ - ▁ex
403
+ - ▁ch
404
+ - ers
405
+ - ▁little
406
+ - ▁actually
407
+ - ▁into
408
+ - ▁make
409
+ - ▁first
410
+ - ▁being
411
+ - ra
412
+ - ▁our
413
+ - ▁al
414
+ - ▁by
415
+ - ▁didn
416
+ - ▁v
417
+ - ct
418
+ - ity
419
+ - ch
420
+ - un
421
+ - ▁part
422
+ - ▁de
423
+ - is
424
+ - ▁film
425
+ - ie
426
+ - ▁right
427
+ - ▁pro
428
+ - ▁off
429
+ - ol
430
+ - ▁two
431
+ - ▁never
432
+ - ▁o
433
+ - ▁
434
+ - ▁le
435
+ - ot
436
+ - ut
437
+ - ▁movie
438
+ - ▁play
439
+ - ge
440
+ - ies
441
+ - el
442
+ - ▁going
443
+ - ke
444
+ - ▁want
445
+ - ▁con
446
+ - ck
447
+ - ▁feel
448
+ - ive
449
+ - ro
450
+ - ▁mo
451
+ - im
452
+ - ▁different
453
+ - ▁life
454
+ - ci
455
+ - am
456
+ - ▁oh
457
+ - all
458
+ - ▁lo
459
+ - ard
460
+ - ▁went
461
+ - and
462
+ - ist
463
+ - ▁sh
464
+ - ▁even
465
+ - ry
466
+ - ▁years
467
+ - ▁look
468
+ - ▁k
469
+ - ▁us
470
+ - ant
471
+ - ▁te
472
+ - ▁li
473
+ - ▁happen
474
+ - ure
475
+ - ▁their
476
+ - ▁those
477
+ - ▁take
478
+ - ment
479
+ - ▁day
480
+ - ast
481
+ - ▁every
482
+ - ill
483
+ - ▁thought
484
+ - ou
485
+ - us
486
+ - ▁th
487
+ - ay
488
+ - ▁put
489
+ - ▁story
490
+ - ▁new
491
+ - ▁down
492
+ - ish
493
+ - ▁big
494
+ - ▁wanna
495
+ - red
496
+ - ▁ro
497
+ - ▁also
498
+ - ▁read
499
+ - ▁around
500
+ - ous
501
+ - ▁through
502
+ - ▁came
503
+ - ▁character
504
+ - ess
505
+ - te
506
+ - ver
507
+ - ▁will
508
+ - ag
509
+ - ss
510
+ - ▁fun
511
+ - ▁over
512
+ - ▁many
513
+ - ▁bl
514
+ - ▁cl
515
+ - ▁man
516
+ - ▁than
517
+ - ▁pre
518
+ - ▁world
519
+ - ▁person
520
+ - z
521
+ - ▁sp
522
+ - ven
523
+ - ▁wanted
524
+ - ▁bit
525
+ - ▁before
526
+ - ▁mar
527
+ - one
528
+ - ab
529
+ - ain
530
+ - ▁en
531
+ - ▁set
532
+ - ▁ha
533
+ - ▁find
534
+ - ul
535
+ - ▁end
536
+ - ▁un
537
+ - ▁sc
538
+ - ▁after
539
+ - een
540
+ - ▁working
541
+ - ▁why
542
+ - ter
543
+ - me
544
+ - ▁such
545
+ - ne
546
+ - ▁whole
547
+ - om
548
+ - ▁kinda
549
+ - pe
550
+ - ▁bo
551
+ - ▁fi
552
+ - x
553
+ - ▁most
554
+ - ▁ad
555
+ - ▁guy
556
+ - ▁spe
557
+ - ars
558
+ - op
559
+ - ▁am
560
+ - ful
561
+ - pt
562
+ - ▁together
563
+ - ▁let
564
+ - ▁quite
565
+ - ▁everything
566
+ - ▁made
567
+ - ig
568
+ - ▁old
569
+ - able
570
+ - ▁comp
571
+ - ▁tr
572
+ - ak
573
+ - ▁fo
574
+ - ▁po
575
+ - ore
576
+ - ice
577
+ - ▁real
578
+ - ▁bas
579
+ - ▁knew
580
+ - ▁hard
581
+ - pp
582
+ - age
583
+ - ated
584
+ - ▁same
585
+ - ▁start
586
+ - ▁ever
587
+ - ning
588
+ - ▁watch
589
+ - art
590
+ - ▁again
591
+ - ▁here
592
+ - are
593
+ - ght
594
+ - ong
595
+ - ▁done
596
+ - ▁only
597
+ - ▁live
598
+ - ▁wasn
599
+ - ▁ho
600
+ - ▁u
601
+ - ▁maybe
602
+ - ▁need
603
+ - ▁everybody
604
+ - ust
605
+ - ▁three
606
+ - ▁having
607
+ - ▁music
608
+ - ack
609
+ - ld
610
+ - ▁trying
611
+ - ▁guys
612
+ - rou
613
+ - ach
614
+ - ving
615
+ - ▁tell
616
+ - ▁should
617
+ - ff
618
+ - ide
619
+ - ▁four
620
+ - ▁started
621
+ - ass
622
+ - ▁long
623
+ - ▁fe
624
+ - ans
625
+ - ▁course
626
+ - ▁called
627
+ - ▁own
628
+ - ress
629
+ - ▁moment
630
+ - ▁pl
631
+ - ▁still
632
+ - ▁anything
633
+ - ▁family
634
+ - ▁fin
635
+ - ▁dan
636
+ - ▁bro
637
+ - 'no'
638
+ - ▁com
639
+ - ther
640
+ - ▁amazing
641
+ - ▁stuff
642
+ - os
643
+ - ▁per
644
+ - ▁jo
645
+ - ▁certain
646
+ - ▁talk
647
+ - ater
648
+ - per
649
+ - ▁help
650
+ - ▁too
651
+ - ▁year
652
+ - ight
653
+ - ▁fa
654
+ - self
655
+ - ces
656
+ - ▁br
657
+ - ▁bet
658
+ - ▁someone
659
+ - ▁di
660
+ - ▁sing
661
+ - nt
662
+ - ick
663
+ - ▁ph
664
+ - row
665
+ - ▁script
666
+ - ▁remember
667
+ - ▁try
668
+ - qu
669
+ - ite
670
+ - ▁young
671
+ - ▁wh
672
+ - ▁ser
673
+ - ▁ask
674
+ - um
675
+ - ▁book
676
+ - ▁each
677
+ - ▁wr
678
+ - ▁best
679
+ - ▁ag
680
+ - ▁women
681
+ - ose
682
+ - ions
683
+ - ved
684
+ - j
685
+ - ue
686
+ - ▁does
687
+ - ty
688
+ - ▁five
689
+ - ▁both
690
+ - ▁friends
691
+ - ▁act
692
+ - iz
693
+ - ind
694
+ - cess
695
+ - ▁somebody
696
+ - ft
697
+ - ▁nice
698
+ - ▁tur
699
+ - ▁myself
700
+ - mb
701
+ - fe
702
+ - ict
703
+ - ▁child
704
+ - ud
705
+ - ▁hope
706
+ - ▁fact
707
+ - ▁saying
708
+ - les
709
+ - ave
710
+ - icul
711
+ - au
712
+ - ris
713
+ - ▁twenty
714
+ - ▁school
715
+ - ▁doesn
716
+ - ▁able
717
+ - pect
718
+ - ▁last
719
+ - ▁song
720
+ - od
721
+ - ▁str
722
+ - ▁interesting
723
+ - lf
724
+ - ▁wor
725
+ - sp
726
+ - ap
727
+ - og
728
+ - ▁ra
729
+ - ▁dis
730
+ - ▁coming
731
+ - ▁ab
732
+ - ▁house
733
+ - ▁next
734
+ - ▁tra
735
+ - ▁okay
736
+ - ere
737
+ - ib
738
+ - ary
739
+ - ▁incredib
740
+ - ▁car
741
+ - ▁job
742
+ - ▁used
743
+ - ▁give
744
+ - ▁god
745
+ - ▁americ
746
+ - ▁characters
747
+ - ▁app
748
+ - ▁walk
749
+ - ▁yes
750
+ - rew
751
+ - ▁getting
752
+ - ▁six
753
+ - ▁chan
754
+ - ▁ne
755
+ - ale
756
+ - ▁pretty
757
+ - mp
758
+ - ang
759
+ - ▁creat
760
+ - ▁another
761
+ - ▁ter
762
+ - ▁kids
763
+ - ▁felt
764
+ - ▁sometimes
765
+ - ▁place
766
+ - ▁int
767
+ - ically
768
+ - out
769
+ - ▁funny
770
+ - ase
771
+ - ich
772
+ - act
773
+ - ▁days
774
+ - ▁bring
775
+ - ▁making
776
+ - ▁become
777
+ - ute
778
+ - ▁wonderful
779
+ - ron
780
+ - ▁saw
781
+ - ▁point
782
+ - ia
783
+ - ▁realiz
784
+ - ▁away
785
+ - ays
786
+ - ▁home
787
+ - ace
788
+ - ▁relationship
789
+ - day
790
+ - ▁woman
791
+ - ▁everyone
792
+ - ▁comes
793
+ - ▁high
794
+ - ▁wee
795
+ - dd
796
+ - ▁night
797
+ - ath
798
+ - ts
799
+ - ▁else
800
+ - vent
801
+ - ▁shoot
802
+ - vers
803
+ - ▁sure
804
+ - ried
805
+ - ned
806
+ - ▁obviously
807
+ - ▁dra
808
+ - co
809
+ - iew
810
+ - man
811
+ - ▁playing
812
+ - ▁important
813
+ - ort
814
+ - uck
815
+ - ision
816
+ - pport
817
+ - ▁nor
818
+ - ▁seen
819
+ - ▁fl
820
+ - est
821
+ - ▁inter
822
+ - ks
823
+ - ▁actor
824
+ - ▁lear
825
+ - ▁worked
826
+ - ▁believe
827
+ - ▁gen
828
+ - ▁keep
829
+ - ull
830
+ - ▁friend
831
+ - ▁sw
832
+ - ▁des
833
+ - ▁times
834
+ - ▁sur
835
+ - ms
836
+ - ▁sit
837
+ - ▁probably
838
+ - ok
839
+ - ▁took
840
+ - ep
841
+ - ough
842
+ - ip
843
+ - ood
844
+ - ▁sa
845
+ - ▁season
846
+ - vel
847
+ - wn
848
+ - ▁dec
849
+ - ▁excited
850
+ - ame
851
+ - ian
852
+ - ire
853
+ - ▁name
854
+ - ▁im
855
+ - ▁month
856
+ - ner
857
+ - ▁min
858
+ - ▁rel
859
+ - ating
860
+ - body
861
+ - ition
862
+ - ▁loved
863
+ - ▁aw
864
+ - ▁hear
865
+ - ph
866
+ - ▁cool
867
+ - ▁list
868
+ - ord
869
+ - pl
870
+ - ble
871
+ - our
872
+ - ▁game
873
+ - ub
874
+ - ▁might
875
+ - ▁kid
876
+ - ▁movies
877
+ - ical
878
+ - ▁bad
879
+ - ▁scene
880
+ - iv
881
+ - ▁enough
882
+ - ▁sm
883
+ - ▁fift
884
+ - ▁eight
885
+ - ▁experience
886
+ - ▁actors
887
+ - ▁understand
888
+ - ▁few
889
+ - gin
890
+ - ting
891
+ - ▁director
892
+ - ▁almost
893
+ - ▁open
894
+ - ren
895
+ - ▁star
896
+ - ▁room
897
+ - ▁call
898
+ - oy
899
+ - ▁goes
900
+ - ▁told
901
+ - ▁once
902
+ - ▁found
903
+ - arly
904
+ - ations
905
+ - ward
906
+ - ▁audience
907
+ - ird
908
+ - ▁qu
909
+ - ▁ar
910
+ - ▁definitely
911
+ - ious
912
+ - iting
913
+ - ▁pol
914
+ - ▁huge
915
+ - ▁makes
916
+ - aking
917
+ - ▁la
918
+ - ▁ac
919
+ - iter
920
+ - ▁run
921
+ - ▁gotta
922
+ - ▁gr
923
+ - ▁cam
924
+ - sh
925
+ - ▁gets
926
+ - ▁wa
927
+ - ully
928
+ - ▁says
929
+ - ▁cont
930
+ - side
931
+ - ▁bus
932
+ - ▁shows
933
+ - ▁dr
934
+ - ▁inv
935
+ - ▁idea
936
+ - ▁talking
937
+ - way
938
+ - ▁art
939
+ - ▁whatever
940
+ - ▁write
941
+ - ash
942
+ - itt
943
+ - ▁met
944
+ - ▁wants
945
+ - ▁role
946
+ - if
947
+ - ▁mu
948
+ - ▁boy
949
+ - ▁wrote
950
+ - ger
951
+ - ately
952
+ - ▁exc
953
+ - ▁gu
954
+ - ▁mother
955
+ - ▁produ
956
+ - ▁cra
957
+ - ates
958
+ - ▁though
959
+ - av
960
+ - ▁episode
961
+ - ▁sl
962
+ - ▁change
963
+ - be
964
+ - ▁voice
965
+ - ▁played
966
+ - ily
967
+ - ▁guess
968
+ - ves
969
+ - ▁hand
970
+ - ady
971
+ - ▁happy
972
+ - ith
973
+ - ny
974
+ - ▁gi
975
+ - med
976
+ - ▁looking
977
+ - lev
978
+ - ream
979
+ - ▁acting
980
+ - aught
981
+ - iss
982
+ - ount
983
+ - rom
984
+ - ▁tw
985
+ - ▁john
986
+ - ▁far
987
+ - ▁res
988
+ - ▁sense
989
+ - ake
990
+ - ▁meet
991
+ - ▁bre
992
+ - ens
993
+ - ety
994
+ - ▁girl
995
+ - ▁york
996
+ - ▁count
997
+ - ▁shot
998
+ - ise
999
+ - ject
1000
+ - ▁tot
1001
+ - ▁stud
1002
+ - ▁feels
1003
+ - ▁thinking
1004
+ - ma
1005
+ - ▁head
1006
+ - ▁cast
1007
+ - ▁writing
1008
+ - ▁imp
1009
+ - ▁rehe
1010
+ - ▁written
1011
+ - ▁perfor
1012
+ - ▁fan
1013
+ - der
1014
+ - ect
1015
+ - ▁sk
1016
+ - ▁hour
1017
+ - ▁father
1018
+ - ered
1019
+ - ▁hundred
1020
+ - ▁ind
1021
+ - ▁che
1022
+ - ▁acc
1023
+ - up
1024
+ - ▁while
1025
+ - fort
1026
+ - ▁true
1027
+ - itch
1028
+ - ▁inst
1029
+ - ▁second
1030
+ - ▁pick
1031
+ - ▁record
1032
+ - ross
1033
+ - ▁quest
1034
+ - ged
1035
+ - ▁career
1036
+ - ▁reason
1037
+ - ▁since
1038
+ - ▁bu
1039
+ - ▁bra
1040
+ - ▁char
1041
+ - ree
1042
+ - ▁girls
1043
+ - ▁dad
1044
+ - ▁fant
1045
+ - ▁extra
1046
+ - ▁laugh
1047
+ - ▁stand
1048
+ - ▁honest
1049
+ - na
1050
+ - als
1051
+ - ▁yet
1052
+ - ▁human
1053
+ - ▁couple
1054
+ - dy
1055
+ - ▁mind
1056
+ - ▁sound
1057
+ - ▁ke
1058
+ - ▁pop
1059
+ - ▁ent
1060
+ - ory
1061
+ - ▁war
1062
+ - ▁ten
1063
+ - ink
1064
+ - ▁bec
1065
+ - ▁direct
1066
+ - reat
1067
+ - round
1068
+ - ien
1069
+ - ▁under
1070
+ - ile
1071
+ - ▁diff
1072
+ - ually
1073
+ - thing
1074
+ - sic
1075
+ - ▁gon
1076
+ - ather
1077
+ - ▁aud
1078
+ - ert
1079
+ - for
1080
+ - ▁scen
1081
+ - mber
1082
+ - atch
1083
+ - ▁sho
1084
+ - ever
1085
+ - tra
1086
+ - ▁pe
1087
+ - ▁hu
1088
+ - ild
1089
+ - int
1090
+ - ▁ob
1091
+ - ▁care
1092
+ - ▁fam
1093
+ - ▁ide
1094
+ - ade
1095
+ - right
1096
+ - ▁may
1097
+ - he
1098
+ - mo
1099
+ - ody
1100
+ - ense
1101
+ - ▁interest
1102
+ - ah
1103
+ - ork
1104
+ - ▁episod
1105
+ - ▁prob
1106
+ - ▁rec
1107
+ - ▁hop
1108
+ - ited
1109
+ - ▁exper
1110
+ - gh
1111
+ - ▁bel
1112
+ - ▁el
1113
+ - ▁stu
1114
+ - enty
1115
+ - ound
1116
+ - ▁gott
1117
+ - ▁id
1118
+ - ime
1119
+ - rie
1120
+ - ▁inc
1121
+ - ertain
1122
+ - ▁wo
1123
+ - ▁mon
1124
+ - az
1125
+ - xt
1126
+ - riend
1127
+ - now
1128
+ - ▁y
1129
+ - ple
1130
+ - ome
1131
+ - so
1132
+ - ause
1133
+ - ▁cou
1134
+ - iously
1135
+ - ▁sch
1136
+ - ▁vo
1137
+ - ▁fil
1138
+ - ▁op
1139
+ - ason
1140
+ - ▁mov
1141
+ - ▁hi
1142
+ - ▁pers
1143
+ - ▁ye
1144
+ - ▁def
1145
+ - ▁belie
1146
+ - fore
1147
+ - ix
1148
+ - very
1149
+ - ▁differe
1150
+ - ▁wonder
1151
+ - nder
1152
+ - ▁obv
1153
+ - ▁ep
1154
+ - ship
1155
+ - ▁lau
1156
+ - ience
1157
+ - ool
1158
+ - ▁sin
1159
+ - rect
1160
+ - ▁happ
1161
+ - ▁gir
1162
+ - ▁hel
1163
+ - du
1164
+ - ng
1165
+ - ▁underst
1166
+ - most
1167
+ - eric
1168
+ - ouse
1169
+ - time
1170
+ - ▁cour
1171
+ - ▁relation
1172
+ - rough
1173
+ - q
1174
+ - ▁defin
1175
+ - ▁reme
1176
+ - redib
1177
+ - ▁fir
1178
+ - anna
1179
+ - ways
1180
+ - itten
1181
+ - elt
1182
+ - ▁sometime
1183
+ - ':'
1184
+ - alk
1185
+ - ▁ok
1186
+ - ably
1187
+ - rote
1188
+ - gether
1189
+ - ▁definite
1190
+ - ▁import
1191
+ - '&'
1192
+ - new
1193
+ - fter
1194
+ - onest
1195
+ - erest
1196
+ - ▁amaz
1197
+ - ▁ano
1198
+ - <sos/eos>
1199
+ transcript_token_list: null
1200
+ two_pass: false
1201
+ pre_postencoder_norm: false
1202
+ init: null
1203
+ input_size: null
1204
+ ctc_conf:
1205
+ dropout_rate: 0.0
1206
+ ctc_type: builtin
1207
+ reduce: true
1208
+ ignore_nan_grad: null
1209
+ zero_infinity: true
1210
+ brctc_risk_strategy: exp
1211
+ brctc_group_strategy: end
1212
+ brctc_risk_factor: 0.0
1213
+ joint_net_conf: null
1214
+ use_preprocessor: true
1215
+ token_type: word
1216
+ bpemodel: null
1217
+ non_linguistic_symbols: null
1218
+ cleaner: null
1219
+ g2p: null
1220
+ speech_volume_normalize: null
1221
+ rir_scp: null
1222
+ rir_apply_prob: 1.0
1223
+ noise_scp: null
1224
+ noise_apply_prob: 1.0
1225
+ noise_db_range: '13_15'
1226
+ short_noise_thres: 0.5
1227
+ frontend: default
1228
+ frontend_conf:
1229
+ n_fft: 512
1230
+ win_length: 400
1231
+ hop_length: 160
1232
+ fs: 16k
1233
+ specaug: specaug
1234
+ specaug_conf:
1235
+ apply_time_warp: false
1236
+ time_warp_window: 5
1237
+ time_warp_mode: bicubic
1238
+ apply_freq_mask: true
1239
+ freq_mask_width_range:
1240
+ - 0
1241
+ - 27
1242
+ num_freq_mask: 2
1243
+ apply_time_mask: true
1244
+ time_mask_width_ratio_range:
1245
+ - 0.0
1246
+ - 0.05
1247
+ num_time_mask: 10
1248
+ normalize: global_mvn
1249
+ normalize_conf:
1250
+ stats_file: /scratch/bbjs/arora1/new_download_espnet_egs2/harpervalley/slu1_superb_onlyda/owsm_v3.1_ebf/exp/s2t_stats_raw_bpe50000/train/feats_stats.npz
1251
+ model: espnet
1252
+ model_conf:
1253
+ ctc_weight: 1.0
1254
+ lsm_weight: 0.1
1255
+ length_normalized_loss: false
1256
+ weighted_sum: true
1257
+ extract_feats_in_collect_stats: false
1258
+ preencoder: null
1259
+ preencoder_conf: {}
1260
+ encoder: e_branchformer
1261
+ encoder_conf:
1262
+ output_size: 1024
1263
+ attention_heads: 16
1264
+ attention_layer_type: selfattn
1265
+ pos_enc_layer_type: abs_pos
1266
+ rel_pos_type: latest
1267
+ cgmlp_linear_units: 4096
1268
+ cgmlp_conv_kernel: 31
1269
+ use_linear_after_conv: false
1270
+ gate_activation: identity
1271
+ num_blocks: 18
1272
+ dropout_rate: 0.1
1273
+ positional_dropout_rate: 0.1
1274
+ attention_dropout_rate: 0.1
1275
+ input_layer: conv2d
1276
+ layer_drop_rate: 0.0
1277
+ linear_units: 4096
1278
+ positionwise_layer_type: linear
1279
+ use_ffn: true
1280
+ macaron_ffn: true
1281
+ merge_conv_kernel: 31
1282
+ prepostencoder: linear
1283
+ prepostencoder_conf:
1284
+ input_size: 1024
1285
+ output_size: 80
1286
+ postencoder: conformer_full
1287
+ postencoder_conf:
1288
+ output_size: 256
1289
+ attention_heads: 4
1290
+ linear_units: 1024
1291
+ num_blocks: 2
1292
+ dropout_rate: 0.1
1293
+ positional_dropout_rate: 0.1
1294
+ attention_dropout_rate: 0.1
1295
+ input_layer: conv2d1
1296
+ normalize_before: true
1297
+ macaron_style: true
1298
+ rel_pos_type: latest
1299
+ pos_enc_layer_type: rel_pos
1300
+ selfattention_layer_type: rel_selfattn
1301
+ activation_type: swish
1302
+ use_cnn_module: true
1303
+ cnn_module_kernel: 31
1304
+ deliberationencoder: null
1305
+ deliberationencoder_conf: {}
1306
+ decoder: transformer
1307
+ decoder_conf:
1308
+ attention_heads: 4
1309
+ linear_units: 2048
1310
+ num_blocks: 6
1311
+ dropout_rate: 0.1
1312
+ positional_dropout_rate: 0.1
1313
+ self_attention_dropout_rate: 0.1
1314
+ src_attention_dropout_rate: 0.1
1315
+ postdecoder: null
1316
+ postdecoder_conf: {}
1317
+ required:
1318
+ - output_dir
1319
+ - token_list
1320
+ version: '202310'
1321
+ distributed: true
1322
+ ```
1323
+
1324
+ </details>
1325
+
1326
+
1327
+
1328
+ ### Citing ESPnet
1329
+
1330
+ ```BibTex
1331
+ @inproceedings{watanabe2018espnet,
1332
+ 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},
1333
+ title={{ESPnet}: End-to-End Speech Processing Toolkit},
1334
+ year={2018},
1335
+ booktitle={Proceedings of Interspeech},
1336
+ pages={2207--2211},
1337
+ doi={10.21437/Interspeech.2018-1456},
1338
+ url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
1339
+ }
1340
+
1341
+
1342
+
1343
+
1344
+
1345
+
1346
+ ```
1347
+
1348
+ or arXiv:
1349
+
1350
+ ```bibtex
1351
+ @misc{watanabe2018espnet,
1352
+ title={ESPnet: End-to-End Speech Processing Toolkit},
1353
+ 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},
1354
+ year={2018},
1355
+ eprint={1804.00015},
1356
+ archivePrefix={arXiv},
1357
+ primaryClass={cs.CL}
1358
+ }
1359
+ ```
exp/slu_train_asr_owsm_superb_raw_en_word_sp/RESULTS.md ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!-- Generated by scripts/utils/show_asr_result.sh -->
2
+ # RESULTS
3
+ ## Environments
4
+ - date: `Mon Feb 5 15:08:12 CST 2024`
5
+ - python version: `3.9.13 (main, Aug 25 2022, 23:26:10) [GCC 11.2.0]`
6
+ - espnet version: `espnet 202310`
7
+ - pytorch version: `pytorch 2.1.0+cu121`
8
+ - Git hash: `21d2105784e4da98397bf487b2550d4c6e16d40d`
9
+ - Commit date: `Wed Jan 31 13:40:37 2024 -0600`
10
+
11
+ ## exp/slu_train_asr_owsm_superb_raw_en_word_sp
12
+ ### WER
13
+
14
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
15
+ |---|---|---|---|---|---|---|---|---|
16
+ |decode_asr_ctc_slu_model_valid.cer_ctc.ave/test|3426|135368|87.2|7.7|5.1|3.5|16.3|93.9|
17
+
18
+ ### CER
19
+
20
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
21
+ |---|---|---|---|---|---|---|---|---|
22
+ |decode_asr_ctc_slu_model_valid.cer_ctc.ave/test|3426|591261|93.9|1.8|4.4|3.3|9.5|93.9|
23
+
24
+ ### TER
25
+
26
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
27
+ |---|---|---|---|---|---|---|---|---|
28
+ ## exp/slu_train_asr_owsm_superb_raw_en_word_sp/decode_asr_ctc_slu_model_valid.cer_ctc.ave
29
+ ### WER
30
+
31
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
32
+ |---|---|---|---|---|---|---|---|---|
33
+ |org/devel|1437|56031|89.2|6.3|4.4|3.1|13.9|92.6|
34
+
35
+ ### CER
36
+
37
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
38
+ |---|---|---|---|---|---|---|---|---|
39
+ |org/devel|1437|241556|95.0|1.3|3.6|2.9|7.9|92.6|
40
+
41
+ ### TER
42
+
43
+ |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
44
+ |---|---|---|---|---|---|---|---|---|
exp/slu_train_asr_owsm_superb_raw_en_word_sp/config.yaml ADDED
@@ -0,0 +1,1241 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ config: conf/tuning/train_asr_owsm_superb.yaml
2
+ print_config: false
3
+ log_level: INFO
4
+ drop_last_iter: false
5
+ dry_run: false
6
+ iterator_type: sequence
7
+ valid_iterator_type: null
8
+ output_dir: exp/slu_train_asr_owsm_superb_raw_en_word_sp
9
+ ngpu: 1
10
+ seed: 2022
11
+ num_workers: 2
12
+ num_att_plot: 3
13
+ dist_backend: nccl
14
+ dist_init_method: env://
15
+ dist_world_size: 4
16
+ dist_rank: 0
17
+ local_rank: 0
18
+ dist_master_addr: localhost
19
+ dist_master_port: 48677
20
+ dist_launcher: null
21
+ multiprocessing_distributed: true
22
+ unused_parameters: false
23
+ sharded_ddp: false
24
+ cudnn_enabled: true
25
+ cudnn_benchmark: false
26
+ cudnn_deterministic: true
27
+ collect_stats: false
28
+ write_collected_feats: false
29
+ max_epoch: 70
30
+ patience: null
31
+ val_scheduler_criterion:
32
+ - valid
33
+ - loss
34
+ early_stopping_criterion:
35
+ - valid
36
+ - loss
37
+ - min
38
+ best_model_criterion:
39
+ - - valid
40
+ - cer_ctc
41
+ - min
42
+ - - valid
43
+ - loss
44
+ - min
45
+ keep_nbest_models: 10
46
+ nbest_averaging_interval: 0
47
+ grad_clip: 5.0
48
+ grad_clip_type: 2.0
49
+ grad_noise: false
50
+ accum_grad: 2
51
+ no_forward_run: false
52
+ resume: true
53
+ train_dtype: float32
54
+ use_amp: false
55
+ log_interval: null
56
+ use_matplotlib: true
57
+ use_tensorboard: true
58
+ create_graph_in_tensorboard: false
59
+ use_wandb: false
60
+ wandb_project: null
61
+ wandb_id: null
62
+ wandb_entity: null
63
+ wandb_name: null
64
+ wandb_model_log_interval: -1
65
+ detect_anomaly: false
66
+ use_lora: false
67
+ save_lora_only: true
68
+ lora_conf: {}
69
+ pretrain_path: null
70
+ init_param:
71
+ - /scratch/bbjs/arora1/new_download_espnet_egs2/harpervalley/slu1_superb_onlyda/owsm_v3.1_ebf/exp/s2t_train_s2t_ebf_conv2d_size1024_e18_d18_piecewise_lr2e-4_warmup60k_flashattn_raw_bpe50000/valid.total_count.ave_5best.till45epoch.pth:encoder:encoder
72
+ ignore_init_mismatch: false
73
+ freeze_param:
74
+ - encoder
75
+ num_iters_per_epoch: null
76
+ batch_size: 20
77
+ valid_batch_size: null
78
+ batch_bins: 6000000
79
+ valid_batch_bins: null
80
+ train_shape_file:
81
+ - exp/slu_stats_raw_en_word_sp/train/speech_shape
82
+ - exp/slu_stats_raw_en_word_sp/train/text_shape.word
83
+ valid_shape_file:
84
+ - exp/slu_stats_raw_en_word_sp/valid/speech_shape
85
+ - exp/slu_stats_raw_en_word_sp/valid/text_shape.word
86
+ batch_type: numel
87
+ valid_batch_type: null
88
+ fold_length:
89
+ - 80000
90
+ - 150
91
+ sort_in_batch: descending
92
+ shuffle_within_batch: false
93
+ sort_batch: descending
94
+ multiple_iterator: false
95
+ chunk_length: 500
96
+ chunk_shift_ratio: 0.5
97
+ num_cache_chunks: 1024
98
+ chunk_excluded_key_prefixes: []
99
+ chunk_default_fs: null
100
+ train_data_path_and_name_and_type:
101
+ - - dump/raw/train_sp/wav.scp
102
+ - speech
103
+ - sound
104
+ - - dump/raw/train_sp/text
105
+ - text
106
+ - text
107
+ valid_data_path_and_name_and_type:
108
+ - - dump/raw/devel/wav.scp
109
+ - speech
110
+ - sound
111
+ - - dump/raw/devel/text
112
+ - text
113
+ - text
114
+ allow_variable_data_keys: false
115
+ max_cache_size: 0.0
116
+ max_cache_fd: 32
117
+ allow_multi_rates: false
118
+ valid_max_cache_size: null
119
+ exclude_weight_decay: false
120
+ exclude_weight_decay_conf: {}
121
+ optim: adam
122
+ optim_conf:
123
+ lr: 0.01
124
+ weight_decay: 1.0e-06
125
+ scheduler: warmuplr
126
+ scheduler_conf:
127
+ warmup_steps: 5000
128
+ token_list:
129
+ - <blank>
130
+ - <unk>
131
+ - ▁i
132
+ - ▁and
133
+ - ''''
134
+ - s
135
+ - ▁the
136
+ - ▁a
137
+ - ▁it
138
+ - ▁to
139
+ - ▁you
140
+ - ▁that
141
+ - ▁of
142
+ - ▁in
143
+ - ▁was
144
+ - ▁uh
145
+ - ▁know
146
+ - t
147
+ - ▁so
148
+ - ▁we
149
+ - ▁he
150
+ - ing
151
+ - m
152
+ - ▁um
153
+ - ▁like
154
+ - ed
155
+ - ▁is
156
+ - ▁but
157
+ - ▁just
158
+ - ▁they
159
+ - re
160
+ - y
161
+ - ▁this
162
+ - ▁for
163
+ - ▁be
164
+ - ▁my
165
+ - er
166
+ - ▁with
167
+ - ▁on
168
+ - ▁think
169
+ - ▁have
170
+ - ▁p
171
+ - ▁she
172
+ - ▁me
173
+ - e
174
+ - ▁really
175
+ - ▁there
176
+ - ▁what
177
+ - al
178
+ - ▁m
179
+ - ▁do
180
+ - ▁all
181
+ - a
182
+ - ve
183
+ - ▁as
184
+ - c
185
+ - n
186
+ - ▁about
187
+ - ▁not
188
+ - i
189
+ - ▁at
190
+ - l
191
+ - ▁t
192
+ - ▁had
193
+ - ▁when
194
+ - ▁c
195
+ - g
196
+ - in
197
+ - ▁b
198
+ - d
199
+ - le
200
+ - en
201
+ - ▁out
202
+ - u
203
+ - ly
204
+ - ▁an
205
+ - or
206
+ - ▁people
207
+ - ar
208
+ - ll
209
+ - o
210
+ - ▁are
211
+ - ▁very
212
+ - ▁because
213
+ - es
214
+ - ▁can
215
+ - ▁don
216
+ - ▁s
217
+ - ▁or
218
+ - ▁up
219
+ - it
220
+ - b
221
+ - ▁e
222
+ - ▁one
223
+ - an
224
+ - st
225
+ - ▁if
226
+ - ▁f
227
+ - ▁were
228
+ - p
229
+ - ▁mean
230
+ - ▁d
231
+ - ▁who
232
+ - ▁then
233
+ - ic
234
+ - 'on'
235
+ - ▁no
236
+ - ▁go
237
+ - ▁her
238
+ - ▁g
239
+ - ▁st
240
+ - ▁kind
241
+ - ri
242
+ - ▁would
243
+ - ▁get
244
+ - at
245
+ - r
246
+ - ▁time
247
+ - v
248
+ - ent
249
+ - ▁re
250
+ - h
251
+ - ▁from
252
+ - ▁l
253
+ - ▁said
254
+ - ▁w
255
+ - ▁him
256
+ - ▁how
257
+ - ▁well
258
+ - ▁h
259
+ - ▁gonna
260
+ - ▁lot
261
+ - ▁see
262
+ - w
263
+ - ▁his
264
+ - ce
265
+ - ion
266
+ - ▁been
267
+ - f
268
+ - ▁great
269
+ - ▁yeah
270
+ - ▁love
271
+ - ▁which
272
+ - ▁got
273
+ - k
274
+ - ▁them
275
+ - ▁way
276
+ - ▁n
277
+ - id
278
+ - ▁show
279
+ - ▁some
280
+ - ▁your
281
+ - ▁did
282
+ - ▁sort
283
+ - et
284
+ - ▁has
285
+ - ▁things
286
+ - ▁back
287
+ - ▁where
288
+ - ▁something
289
+ - ir
290
+ - ▁thing
291
+ - ad
292
+ - ▁su
293
+ - il
294
+ - as
295
+ - ▁j
296
+ - ▁more
297
+ - ▁co
298
+ - se
299
+ - ▁say
300
+ - nd
301
+ - ▁much
302
+ - ▁come
303
+ - ▁always
304
+ - ine
305
+ - ▁r
306
+ - ation
307
+ - ▁other
308
+ - th
309
+ - ur
310
+ - ▁se
311
+ - ▁now
312
+ - ate
313
+ - ▁doing
314
+ - ▁work
315
+ - ow
316
+ - ▁could
317
+ - ally
318
+ - ▁these
319
+ - ▁good
320
+ - ▁any
321
+ - ▁cause
322
+ - ▁ex
323
+ - ▁ch
324
+ - ers
325
+ - ▁little
326
+ - ▁actually
327
+ - ▁into
328
+ - ▁make
329
+ - ▁first
330
+ - ▁being
331
+ - ra
332
+ - ▁our
333
+ - ▁al
334
+ - ▁by
335
+ - ▁didn
336
+ - ▁v
337
+ - ct
338
+ - ity
339
+ - ch
340
+ - un
341
+ - ▁part
342
+ - ▁de
343
+ - is
344
+ - ▁film
345
+ - ie
346
+ - ▁right
347
+ - ▁pro
348
+ - ▁off
349
+ - ol
350
+ - ▁two
351
+ - ▁never
352
+ - ▁o
353
+ - ▁
354
+ - ▁le
355
+ - ot
356
+ - ut
357
+ - ▁movie
358
+ - ▁play
359
+ - ge
360
+ - ies
361
+ - el
362
+ - ▁going
363
+ - ke
364
+ - ▁want
365
+ - ▁con
366
+ - ck
367
+ - ▁feel
368
+ - ive
369
+ - ro
370
+ - ▁mo
371
+ - im
372
+ - ▁different
373
+ - ▁life
374
+ - ci
375
+ - am
376
+ - ▁oh
377
+ - all
378
+ - ▁lo
379
+ - ard
380
+ - ▁went
381
+ - and
382
+ - ist
383
+ - ▁sh
384
+ - ▁even
385
+ - ry
386
+ - ▁years
387
+ - ▁look
388
+ - ▁k
389
+ - ▁us
390
+ - ant
391
+ - ▁te
392
+ - ▁li
393
+ - ▁happen
394
+ - ure
395
+ - ▁their
396
+ - ▁those
397
+ - ▁take
398
+ - ment
399
+ - ▁day
400
+ - ast
401
+ - ▁every
402
+ - ill
403
+ - ▁thought
404
+ - ou
405
+ - us
406
+ - ▁th
407
+ - ay
408
+ - ▁put
409
+ - ▁story
410
+ - ▁new
411
+ - ▁down
412
+ - ish
413
+ - ▁big
414
+ - ▁wanna
415
+ - red
416
+ - ▁ro
417
+ - ▁also
418
+ - ▁read
419
+ - ▁around
420
+ - ous
421
+ - ▁through
422
+ - ▁came
423
+ - ▁character
424
+ - ess
425
+ - te
426
+ - ver
427
+ - ▁will
428
+ - ag
429
+ - ss
430
+ - ▁fun
431
+ - ▁over
432
+ - ▁many
433
+ - ▁bl
434
+ - ▁cl
435
+ - ▁man
436
+ - ▁than
437
+ - ▁pre
438
+ - ▁world
439
+ - ▁person
440
+ - z
441
+ - ▁sp
442
+ - ven
443
+ - ▁wanted
444
+ - ▁bit
445
+ - ▁before
446
+ - ▁mar
447
+ - one
448
+ - ab
449
+ - ain
450
+ - ▁en
451
+ - ▁set
452
+ - ▁ha
453
+ - ▁find
454
+ - ul
455
+ - ▁end
456
+ - ▁un
457
+ - ▁sc
458
+ - ▁after
459
+ - een
460
+ - ▁working
461
+ - ▁why
462
+ - ter
463
+ - me
464
+ - ▁such
465
+ - ne
466
+ - ▁whole
467
+ - om
468
+ - ▁kinda
469
+ - pe
470
+ - ▁bo
471
+ - ▁fi
472
+ - x
473
+ - ▁most
474
+ - ▁ad
475
+ - ▁guy
476
+ - ▁spe
477
+ - ars
478
+ - op
479
+ - ▁am
480
+ - ful
481
+ - pt
482
+ - ▁together
483
+ - ▁let
484
+ - ▁quite
485
+ - ▁everything
486
+ - ▁made
487
+ - ig
488
+ - ▁old
489
+ - able
490
+ - ▁comp
491
+ - ▁tr
492
+ - ak
493
+ - ▁fo
494
+ - ▁po
495
+ - ore
496
+ - ice
497
+ - ▁real
498
+ - ▁bas
499
+ - ▁knew
500
+ - ▁hard
501
+ - pp
502
+ - age
503
+ - ated
504
+ - ▁same
505
+ - ▁start
506
+ - ▁ever
507
+ - ning
508
+ - ▁watch
509
+ - art
510
+ - ▁again
511
+ - ▁here
512
+ - are
513
+ - ght
514
+ - ong
515
+ - ▁done
516
+ - ▁only
517
+ - ▁live
518
+ - ▁wasn
519
+ - ▁ho
520
+ - ▁u
521
+ - ▁maybe
522
+ - ▁need
523
+ - ▁everybody
524
+ - ust
525
+ - ▁three
526
+ - ▁having
527
+ - ▁music
528
+ - ack
529
+ - ld
530
+ - ▁trying
531
+ - ▁guys
532
+ - rou
533
+ - ach
534
+ - ving
535
+ - ▁tell
536
+ - ▁should
537
+ - ff
538
+ - ide
539
+ - ▁four
540
+ - ▁started
541
+ - ass
542
+ - ▁long
543
+ - ▁fe
544
+ - ans
545
+ - ▁course
546
+ - ▁called
547
+ - ▁own
548
+ - ress
549
+ - ▁moment
550
+ - ▁pl
551
+ - ▁still
552
+ - ▁anything
553
+ - ▁family
554
+ - ▁fin
555
+ - ▁dan
556
+ - ▁bro
557
+ - 'no'
558
+ - ▁com
559
+ - ther
560
+ - ▁amazing
561
+ - ▁stuff
562
+ - os
563
+ - ▁per
564
+ - ▁jo
565
+ - ▁certain
566
+ - ▁talk
567
+ - ater
568
+ - per
569
+ - ▁help
570
+ - ▁too
571
+ - ▁year
572
+ - ight
573
+ - ▁fa
574
+ - self
575
+ - ces
576
+ - ▁br
577
+ - ▁bet
578
+ - ▁someone
579
+ - ▁di
580
+ - ▁sing
581
+ - nt
582
+ - ick
583
+ - ▁ph
584
+ - row
585
+ - ▁script
586
+ - ▁remember
587
+ - ▁try
588
+ - qu
589
+ - ite
590
+ - ▁young
591
+ - ▁wh
592
+ - ▁ser
593
+ - ▁ask
594
+ - um
595
+ - ▁book
596
+ - ▁each
597
+ - ▁wr
598
+ - ▁best
599
+ - ▁ag
600
+ - ▁women
601
+ - ose
602
+ - ions
603
+ - ved
604
+ - j
605
+ - ue
606
+ - ▁does
607
+ - ty
608
+ - ▁five
609
+ - ▁both
610
+ - ▁friends
611
+ - ▁act
612
+ - iz
613
+ - ind
614
+ - cess
615
+ - ▁somebody
616
+ - ft
617
+ - ▁nice
618
+ - ▁tur
619
+ - ▁myself
620
+ - mb
621
+ - fe
622
+ - ict
623
+ - ▁child
624
+ - ud
625
+ - ▁hope
626
+ - ▁fact
627
+ - ▁saying
628
+ - les
629
+ - ave
630
+ - icul
631
+ - au
632
+ - ris
633
+ - ▁twenty
634
+ - ▁school
635
+ - ▁doesn
636
+ - ▁able
637
+ - pect
638
+ - ▁last
639
+ - ▁song
640
+ - od
641
+ - ▁str
642
+ - ▁interesting
643
+ - lf
644
+ - ▁wor
645
+ - sp
646
+ - ap
647
+ - og
648
+ - ▁ra
649
+ - ▁dis
650
+ - ▁coming
651
+ - ▁ab
652
+ - ▁house
653
+ - ▁next
654
+ - ▁tra
655
+ - ▁okay
656
+ - ere
657
+ - ib
658
+ - ary
659
+ - ▁incredib
660
+ - ▁car
661
+ - ▁job
662
+ - ▁used
663
+ - ▁give
664
+ - ▁god
665
+ - ▁americ
666
+ - ▁characters
667
+ - ▁app
668
+ - ▁walk
669
+ - ▁yes
670
+ - rew
671
+ - ▁getting
672
+ - ▁six
673
+ - ▁chan
674
+ - ▁ne
675
+ - ale
676
+ - ▁pretty
677
+ - mp
678
+ - ang
679
+ - ▁creat
680
+ - ▁another
681
+ - ▁ter
682
+ - ▁kids
683
+ - ▁felt
684
+ - ▁sometimes
685
+ - ▁place
686
+ - ▁int
687
+ - ically
688
+ - out
689
+ - ▁funny
690
+ - ase
691
+ - ich
692
+ - act
693
+ - ▁days
694
+ - ▁bring
695
+ - ▁making
696
+ - ▁become
697
+ - ute
698
+ - ▁wonderful
699
+ - ron
700
+ - ▁saw
701
+ - ▁point
702
+ - ia
703
+ - ▁realiz
704
+ - ▁away
705
+ - ays
706
+ - ▁home
707
+ - ace
708
+ - ▁relationship
709
+ - day
710
+ - ▁woman
711
+ - ▁everyone
712
+ - ▁comes
713
+ - ▁high
714
+ - ▁wee
715
+ - dd
716
+ - ▁night
717
+ - ath
718
+ - ts
719
+ - ▁else
720
+ - vent
721
+ - ▁shoot
722
+ - vers
723
+ - ▁sure
724
+ - ried
725
+ - ned
726
+ - ▁obviously
727
+ - ▁dra
728
+ - co
729
+ - iew
730
+ - man
731
+ - ▁playing
732
+ - ▁important
733
+ - ort
734
+ - uck
735
+ - ision
736
+ - pport
737
+ - ▁nor
738
+ - ▁seen
739
+ - ▁fl
740
+ - est
741
+ - ▁inter
742
+ - ks
743
+ - ▁actor
744
+ - ▁lear
745
+ - ▁worked
746
+ - ▁believe
747
+ - ▁gen
748
+ - ▁keep
749
+ - ull
750
+ - ▁friend
751
+ - ▁sw
752
+ - ▁des
753
+ - ▁times
754
+ - ▁sur
755
+ - ms
756
+ - ▁sit
757
+ - ▁probably
758
+ - ok
759
+ - ▁took
760
+ - ep
761
+ - ough
762
+ - ip
763
+ - ood
764
+ - ▁sa
765
+ - ▁season
766
+ - vel
767
+ - wn
768
+ - ▁dec
769
+ - ▁excited
770
+ - ame
771
+ - ian
772
+ - ire
773
+ - ▁name
774
+ - ▁im
775
+ - ▁month
776
+ - ner
777
+ - ▁min
778
+ - ▁rel
779
+ - ating
780
+ - body
781
+ - ition
782
+ - ▁loved
783
+ - ▁aw
784
+ - ▁hear
785
+ - ph
786
+ - ▁cool
787
+ - ▁list
788
+ - ord
789
+ - pl
790
+ - ble
791
+ - our
792
+ - ▁game
793
+ - ub
794
+ - ▁might
795
+ - ▁kid
796
+ - ▁movies
797
+ - ical
798
+ - ▁bad
799
+ - ▁scene
800
+ - iv
801
+ - ▁enough
802
+ - ▁sm
803
+ - ▁fift
804
+ - ▁eight
805
+ - ▁experience
806
+ - ▁actors
807
+ - ▁understand
808
+ - ▁few
809
+ - gin
810
+ - ting
811
+ - ▁director
812
+ - ▁almost
813
+ - ▁open
814
+ - ren
815
+ - ▁star
816
+ - ▁room
817
+ - ▁call
818
+ - oy
819
+ - ▁goes
820
+ - ▁told
821
+ - ▁once
822
+ - ▁found
823
+ - arly
824
+ - ations
825
+ - ward
826
+ - ▁audience
827
+ - ird
828
+ - ▁qu
829
+ - ▁ar
830
+ - ▁definitely
831
+ - ious
832
+ - iting
833
+ - ▁pol
834
+ - ▁huge
835
+ - ▁makes
836
+ - aking
837
+ - ▁la
838
+ - ▁ac
839
+ - iter
840
+ - ▁run
841
+ - ▁gotta
842
+ - ▁gr
843
+ - ▁cam
844
+ - sh
845
+ - ▁gets
846
+ - ▁wa
847
+ - ully
848
+ - ▁says
849
+ - ▁cont
850
+ - side
851
+ - ▁bus
852
+ - ▁shows
853
+ - ▁dr
854
+ - ▁inv
855
+ - ▁idea
856
+ - ▁talking
857
+ - way
858
+ - ▁art
859
+ - ▁whatever
860
+ - ▁write
861
+ - ash
862
+ - itt
863
+ - ▁met
864
+ - ▁wants
865
+ - ▁role
866
+ - if
867
+ - ▁mu
868
+ - ▁boy
869
+ - ▁wrote
870
+ - ger
871
+ - ately
872
+ - ▁exc
873
+ - ▁gu
874
+ - ▁mother
875
+ - ▁produ
876
+ - ▁cra
877
+ - ates
878
+ - ▁though
879
+ - av
880
+ - ▁episode
881
+ - ▁sl
882
+ - ▁change
883
+ - be
884
+ - ▁voice
885
+ - ▁played
886
+ - ily
887
+ - ▁guess
888
+ - ves
889
+ - ▁hand
890
+ - ady
891
+ - ▁happy
892
+ - ith
893
+ - ny
894
+ - ▁gi
895
+ - med
896
+ - ▁looking
897
+ - lev
898
+ - ream
899
+ - ▁acting
900
+ - aught
901
+ - iss
902
+ - ount
903
+ - rom
904
+ - ▁tw
905
+ - ▁john
906
+ - ▁far
907
+ - ▁res
908
+ - ▁sense
909
+ - ake
910
+ - ▁meet
911
+ - ▁bre
912
+ - ens
913
+ - ety
914
+ - ▁girl
915
+ - ▁york
916
+ - ▁count
917
+ - ▁shot
918
+ - ise
919
+ - ject
920
+ - ▁tot
921
+ - ▁stud
922
+ - ▁feels
923
+ - ▁thinking
924
+ - ma
925
+ - ▁head
926
+ - ▁cast
927
+ - ▁writing
928
+ - ▁imp
929
+ - ▁rehe
930
+ - ▁written
931
+ - ▁perfor
932
+ - ▁fan
933
+ - der
934
+ - ect
935
+ - ▁sk
936
+ - ▁hour
937
+ - ▁father
938
+ - ered
939
+ - ▁hundred
940
+ - ▁ind
941
+ - ▁che
942
+ - ▁acc
943
+ - up
944
+ - ▁while
945
+ - fort
946
+ - ▁true
947
+ - itch
948
+ - ▁inst
949
+ - ▁second
950
+ - ▁pick
951
+ - ▁record
952
+ - ross
953
+ - ▁quest
954
+ - ged
955
+ - ▁career
956
+ - ▁reason
957
+ - ▁since
958
+ - ▁bu
959
+ - ▁bra
960
+ - ▁char
961
+ - ree
962
+ - ▁girls
963
+ - ▁dad
964
+ - ▁fant
965
+ - ▁extra
966
+ - ▁laugh
967
+ - ▁stand
968
+ - ▁honest
969
+ - na
970
+ - als
971
+ - ▁yet
972
+ - ▁human
973
+ - ▁couple
974
+ - dy
975
+ - ▁mind
976
+ - ▁sound
977
+ - ▁ke
978
+ - ▁pop
979
+ - ▁ent
980
+ - ory
981
+ - ▁war
982
+ - ▁ten
983
+ - ink
984
+ - ▁bec
985
+ - ▁direct
986
+ - reat
987
+ - round
988
+ - ien
989
+ - ▁under
990
+ - ile
991
+ - ▁diff
992
+ - ually
993
+ - thing
994
+ - sic
995
+ - ▁gon
996
+ - ather
997
+ - ▁aud
998
+ - ert
999
+ - for
1000
+ - ▁scen
1001
+ - mber
1002
+ - atch
1003
+ - ▁sho
1004
+ - ever
1005
+ - tra
1006
+ - ▁pe
1007
+ - ▁hu
1008
+ - ild
1009
+ - int
1010
+ - ▁ob
1011
+ - ▁care
1012
+ - ▁fam
1013
+ - ▁ide
1014
+ - ade
1015
+ - right
1016
+ - ▁may
1017
+ - he
1018
+ - mo
1019
+ - ody
1020
+ - ense
1021
+ - ▁interest
1022
+ - ah
1023
+ - ork
1024
+ - ▁episod
1025
+ - ▁prob
1026
+ - ▁rec
1027
+ - ▁hop
1028
+ - ited
1029
+ - ▁exper
1030
+ - gh
1031
+ - ▁bel
1032
+ - ▁el
1033
+ - ▁stu
1034
+ - enty
1035
+ - ound
1036
+ - ▁gott
1037
+ - ▁id
1038
+ - ime
1039
+ - rie
1040
+ - ▁inc
1041
+ - ertain
1042
+ - ▁wo
1043
+ - ▁mon
1044
+ - az
1045
+ - xt
1046
+ - riend
1047
+ - now
1048
+ - ▁y
1049
+ - ple
1050
+ - ome
1051
+ - so
1052
+ - ause
1053
+ - ▁cou
1054
+ - iously
1055
+ - ▁sch
1056
+ - ▁vo
1057
+ - ▁fil
1058
+ - ▁op
1059
+ - ason
1060
+ - ▁mov
1061
+ - ▁hi
1062
+ - ▁pers
1063
+ - ▁ye
1064
+ - ▁def
1065
+ - ▁belie
1066
+ - fore
1067
+ - ix
1068
+ - very
1069
+ - ▁differe
1070
+ - ▁wonder
1071
+ - nder
1072
+ - ▁obv
1073
+ - ▁ep
1074
+ - ship
1075
+ - ▁lau
1076
+ - ience
1077
+ - ool
1078
+ - ▁sin
1079
+ - rect
1080
+ - ▁happ
1081
+ - ▁gir
1082
+ - ▁hel
1083
+ - du
1084
+ - ng
1085
+ - ▁underst
1086
+ - most
1087
+ - eric
1088
+ - ouse
1089
+ - time
1090
+ - ▁cour
1091
+ - ▁relation
1092
+ - rough
1093
+ - q
1094
+ - ▁defin
1095
+ - ▁reme
1096
+ - redib
1097
+ - ▁fir
1098
+ - anna
1099
+ - ways
1100
+ - itten
1101
+ - elt
1102
+ - ▁sometime
1103
+ - ':'
1104
+ - alk
1105
+ - ▁ok
1106
+ - ably
1107
+ - rote
1108
+ - gether
1109
+ - ▁definite
1110
+ - ▁import
1111
+ - '&'
1112
+ - new
1113
+ - fter
1114
+ - onest
1115
+ - erest
1116
+ - ▁amaz
1117
+ - ▁ano
1118
+ - <sos/eos>
1119
+ transcript_token_list: null
1120
+ two_pass: false
1121
+ pre_postencoder_norm: false
1122
+ init: null
1123
+ input_size: null
1124
+ ctc_conf:
1125
+ dropout_rate: 0.0
1126
+ ctc_type: builtin
1127
+ reduce: true
1128
+ ignore_nan_grad: null
1129
+ zero_infinity: true
1130
+ brctc_risk_strategy: exp
1131
+ brctc_group_strategy: end
1132
+ brctc_risk_factor: 0.0
1133
+ joint_net_conf: null
1134
+ use_preprocessor: true
1135
+ token_type: word
1136
+ bpemodel: null
1137
+ non_linguistic_symbols: null
1138
+ cleaner: null
1139
+ g2p: null
1140
+ speech_volume_normalize: null
1141
+ rir_scp: null
1142
+ rir_apply_prob: 1.0
1143
+ noise_scp: null
1144
+ noise_apply_prob: 1.0
1145
+ noise_db_range: '13_15'
1146
+ short_noise_thres: 0.5
1147
+ frontend: default
1148
+ frontend_conf:
1149
+ n_fft: 512
1150
+ win_length: 400
1151
+ hop_length: 160
1152
+ fs: 16k
1153
+ specaug: specaug
1154
+ specaug_conf:
1155
+ apply_time_warp: false
1156
+ time_warp_window: 5
1157
+ time_warp_mode: bicubic
1158
+ apply_freq_mask: true
1159
+ freq_mask_width_range:
1160
+ - 0
1161
+ - 27
1162
+ num_freq_mask: 2
1163
+ apply_time_mask: true
1164
+ time_mask_width_ratio_range:
1165
+ - 0.0
1166
+ - 0.05
1167
+ num_time_mask: 10
1168
+ normalize: global_mvn
1169
+ normalize_conf:
1170
+ stats_file: /scratch/bbjs/arora1/new_download_espnet_egs2/harpervalley/slu1_superb_onlyda/owsm_v3.1_ebf/exp/s2t_stats_raw_bpe50000/train/feats_stats.npz
1171
+ model: espnet
1172
+ model_conf:
1173
+ ctc_weight: 1.0
1174
+ lsm_weight: 0.1
1175
+ length_normalized_loss: false
1176
+ weighted_sum: true
1177
+ extract_feats_in_collect_stats: false
1178
+ preencoder: null
1179
+ preencoder_conf: {}
1180
+ encoder: e_branchformer
1181
+ encoder_conf:
1182
+ output_size: 1024
1183
+ attention_heads: 16
1184
+ attention_layer_type: selfattn
1185
+ pos_enc_layer_type: abs_pos
1186
+ rel_pos_type: latest
1187
+ cgmlp_linear_units: 4096
1188
+ cgmlp_conv_kernel: 31
1189
+ use_linear_after_conv: false
1190
+ gate_activation: identity
1191
+ num_blocks: 18
1192
+ dropout_rate: 0.1
1193
+ positional_dropout_rate: 0.1
1194
+ attention_dropout_rate: 0.1
1195
+ input_layer: conv2d
1196
+ layer_drop_rate: 0.0
1197
+ linear_units: 4096
1198
+ positionwise_layer_type: linear
1199
+ use_ffn: true
1200
+ macaron_ffn: true
1201
+ merge_conv_kernel: 31
1202
+ prepostencoder: linear
1203
+ prepostencoder_conf:
1204
+ input_size: 1024
1205
+ output_size: 80
1206
+ postencoder: conformer_full
1207
+ postencoder_conf:
1208
+ output_size: 256
1209
+ attention_heads: 4
1210
+ linear_units: 1024
1211
+ num_blocks: 2
1212
+ dropout_rate: 0.1
1213
+ positional_dropout_rate: 0.1
1214
+ attention_dropout_rate: 0.1
1215
+ input_layer: conv2d1
1216
+ normalize_before: true
1217
+ macaron_style: true
1218
+ rel_pos_type: latest
1219
+ pos_enc_layer_type: rel_pos
1220
+ selfattention_layer_type: rel_selfattn
1221
+ activation_type: swish
1222
+ use_cnn_module: true
1223
+ cnn_module_kernel: 31
1224
+ deliberationencoder: null
1225
+ deliberationencoder_conf: {}
1226
+ decoder: transformer
1227
+ decoder_conf:
1228
+ attention_heads: 4
1229
+ linear_units: 2048
1230
+ num_blocks: 6
1231
+ dropout_rate: 0.1
1232
+ positional_dropout_rate: 0.1
1233
+ self_attention_dropout_rate: 0.1
1234
+ src_attention_dropout_rate: 0.1
1235
+ postdecoder: null
1236
+ postdecoder_conf: {}
1237
+ required:
1238
+ - output_dir
1239
+ - token_list
1240
+ version: '202310'
1241
+ distributed: true
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/acc.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/backward_time.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/cer.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/cer_ctc.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/clip.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/forward_time.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/gpu_max_cached_mem_GB.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/grad_norm.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/iter_time.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/loss.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/loss_att.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/loss_ctc.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/loss_scale.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/optim0_lr0.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/optim_step_time.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/train_time.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/images/wer.png ADDED
exp/slu_train_asr_owsm_superb_raw_en_word_sp/valid.loss.ave_10best.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:41befd00e6b336736293c1407f4c23a0d927a56c957583b2aa66a890b0838a99
3
+ size 2280115930
meta.yaml ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ espnet: '202310'
2
+ files:
3
+ slu_model_file: exp/slu_train_asr_owsm_superb_raw_en_word_sp/valid.loss.ave_10best.pth
4
+ python: "3.9.13 (main, Aug 25 2022, 23:26:10) \n[GCC 11.2.0]"
5
+ timestamp: 1715351537.536378
6
+ torch: 2.1.0+cu121
7
+ yaml_files:
8
+ slu_train_config: exp/slu_train_asr_owsm_superb_raw_en_word_sp/config.yaml