update tedlium3-transducer-stateless files

README.md

@@ -27,13 +27,13 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3"
                   --num-epochs 30 \
                   --start-epoch 0 \
                   --exp-dir transducer_stateless/exp \
-                  --max-duration 200
+                  --max-duration 300
 ```
 ## Evaluation results
-The decoding results (WER%) on TEDLium3 (dev and test) are listed below, we got this result by averaging models from epoch 14 to 29.
+The decoding results (WER%) on TEDLium3 (dev and test) are listed below, we got this result by averaging models from epoch 19 to 29.
 The WERs are
 |                                    |     dev    |    test    | comment                                  |
 |------------------------------------|------------|------------|------------------------------------------|
-|          greedy search             | 7.19       | 6.57       | --epoch 29, --avg 16, --max-duration 100 |
-|      beam search (beam size 4)     | 7.12       | 6.37       | --epoch 29, --avg 16, --max-duration 100 |
-| modified beam search (beam size 4) | 7.00       | 6.19       | --epoch 29, --avg 16, --max-duration 100 |
+|          greedy search             | 7.19       | 6.70       | --epoch 29, --avg 11, --max-duration 100 |
+|      beam search (beam size 4)     | 7.02       | 6.36       | --epoch 29, --avg 11, --max-duration 100 |
+| modified beam search (beam size 4) | 6.91       | 6.33       | --epoch 29, --avg 11, --max-duration 100 |

exp/pretrained_average_19_to_29.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aae8eef5ad06f8ded4ea842746d4ce686d2c2ab250c4e71e91c761cf3899607f
+size 1008510957

log/beam_search/log-decode-epoch-29-avg-11-beam-4-2022-03-21-15-43-55

@@ -0,0 +1,112 @@
+2022-03-21 15:43:55,820 INFO [decode.py:425] Decoding started
+2022-03-21 15:43:55,909 INFO [decode.py:431] Device: cuda:0
+2022-03-21 15:43:55,912 INFO [decode.py:441] {'feature_dim': 80, 'encoder_out_dim': 512, 'subsampling_factor': 4, 'attention_dim': 512, 'nhead': 8, 'dim_feedforward': 2048, 'num_encoder_layers': 12, 'vgg_frontend': False, 'epoch': 29, 'avg': 11, 'exp_dir': PosixPath('transducer_stateless/exp'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'decoding_method': 'beam_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'res_dir': PosixPath('transducer_stateless/exp/beam_search'), 'suffix': 'epoch-29-avg-11-beam-4', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
+2022-03-21 15:43:55,912 INFO [decode.py:443] About to create model
+2022-03-21 15:43:56,507 INFO [decode.py:454] averaging ['transducer_stateless/exp/epoch-19.pt', 'transducer_stateless/exp/epoch-20.pt', 'transducer_stateless/exp/epoch-21.pt', 'transducer_stateless/exp/epoch-22.pt', 'transducer_stateless/exp/epoch-23.pt', 'transducer_stateless/exp/epoch-24.pt', 'transducer_stateless/exp/epoch-25.pt', 'transducer_stateless/exp/epoch-26.pt', 'transducer_stateless/exp/epoch-27.pt', 'transducer_stateless/exp/epoch-28.pt', 'transducer_stateless/exp/epoch-29.pt']
+2022-03-21 15:44:10,396 INFO [decode.py:463] Number of model parameters: 84007924
+2022-03-21 15:44:10,396 INFO [asr_datamodule.py:357] About to get dev cuts
+2022-03-21 15:44:10,420 INFO [asr_datamodule.py:362] About to get test cuts
+2022-03-21 15:44:10,472 INFO [asr_datamodule.py:300] About to create dev dataset
+2022-03-21 15:44:10,473 INFO [asr_datamodule.py:319] About to create dev dataloader
+2022-03-21 15:44:20,650 INFO [decode.py:350] batch 0/?, cuts processed until now is 10
+2022-03-21 15:44:42,580 INFO [decode.py:350] batch 2/?, cuts processed until now is 33
+2022-03-21 15:45:04,852 INFO [decode.py:350] batch 4/?, cuts processed until now is 45
+2022-03-21 15:45:26,348 INFO [decode.py:350] batch 6/?, cuts processed until now is 67
+2022-03-21 15:45:48,094 INFO [decode.py:350] batch 8/?, cuts processed until now is 77
+2022-03-21 15:46:10,105 INFO [decode.py:350] batch 10/?, cuts processed until now is 96
+2022-03-21 15:46:33,479 INFO [decode.py:350] batch 12/?, cuts processed until now is 101
+2022-03-21 15:46:54,714 INFO [decode.py:350] batch 14/?, cuts processed until now is 111
+2022-03-21 15:47:15,052 INFO [decode.py:350] batch 16/?, cuts processed until now is 125
+2022-03-21 15:47:37,035 INFO [decode.py:350] batch 18/?, cuts processed until now is 140
+2022-03-21 15:47:59,475 INFO [decode.py:350] batch 20/?, cuts processed until now is 158
+2022-03-21 15:48:21,447 INFO [decode.py:350] batch 22/?, cuts processed until now is 184
+2022-03-21 15:48:42,757 INFO [decode.py:350] batch 24/?, cuts processed until now is 198
+2022-03-21 15:49:03,381 INFO [decode.py:350] batch 26/?, cuts processed until now is 209
+2022-03-21 15:49:23,840 INFO [decode.py:350] batch 28/?, cuts processed until now is 223
+2022-03-21 15:49:46,027 INFO [decode.py:350] batch 30/?, cuts processed until now is 237
+2022-03-21 15:50:08,756 INFO [decode.py:350] batch 32/?, cuts processed until now is 256
+2022-03-21 15:50:30,116 INFO [decode.py:350] batch 34/?, cuts processed until now is 278
+2022-03-21 15:50:52,911 INFO [decode.py:350] batch 36/?, cuts processed until now is 294
+2022-03-21 15:51:14,210 INFO [decode.py:350] batch 38/?, cuts processed until now is 306
+2022-03-21 15:51:37,797 INFO [decode.py:350] batch 40/?, cuts processed until now is 314
+2022-03-21 15:51:59,394 INFO [decode.py:350] batch 42/?, cuts processed until now is 345
+2022-03-21 15:52:15,166 INFO [decode.py:350] batch 44/?, cuts processed until now is 378
+2022-03-21 15:52:35,771 INFO [decode.py:350] batch 46/?, cuts processed until now is 388
+2022-03-21 15:52:55,915 INFO [decode.py:350] batch 48/?, cuts processed until now is 404
+2022-03-21 15:53:11,067 INFO [decode.py:350] batch 50/?, cuts processed until now is 412
+2022-03-21 15:53:32,654 INFO [decode.py:350] batch 52/?, cuts processed until now is 425
+2022-03-21 15:53:45,268 INFO [decode.py:350] batch 54/?, cuts processed until now is 435
+2022-03-21 15:54:07,215 INFO [decode.py:350] batch 56/?, cuts processed until now is 446
+2022-03-21 15:54:19,335 INFO [decode.py:350] batch 58/?, cuts processed until now is 458
+2022-03-21 15:54:42,879 INFO [decode.py:350] batch 60/?, cuts processed until now is 474
+2022-03-21 15:54:58,880 INFO [decode.py:350] batch 62/?, cuts processed until now is 483
+2022-03-21 15:55:11,143 INFO [decode.py:350] batch 64/?, cuts processed until now is 493
+2022-03-21 15:55:34,175 INFO [decode.py:350] batch 66/?, cuts processed until now is 507
+2022-03-21 15:55:34,291 INFO [decode.py:367] The transcripts are stored in transducer_stateless/exp/beam_search/recogs-dev-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 15:55:34,318 INFO [utils.py:406] [dev-beam_4] %WER 7.02% [1280 / 18226, 215 ins, 391 del, 674 sub ]
+2022-03-21 15:55:34,380 INFO [decode.py:380] Wrote detailed error stats to transducer_stateless/exp/beam_search/errs-dev-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 15:55:34,381 INFO [decode.py:397] 
+For dev, WER of different settings are:
+beam_4	7.02	best for dev
+
+2022-03-21 15:55:45,328 INFO [decode.py:350] batch 0/?, cuts processed until now is 14
+2022-03-21 15:56:07,644 INFO [decode.py:350] batch 2/?, cuts processed until now is 51
+2022-03-21 15:56:29,901 INFO [decode.py:350] batch 4/?, cuts processed until now is 67
+2022-03-21 15:56:51,937 INFO [decode.py:350] batch 6/?, cuts processed until now is 104
+2022-03-21 15:57:13,580 INFO [decode.py:350] batch 8/?, cuts processed until now is 118
+2022-03-21 15:57:34,769 INFO [decode.py:350] batch 10/?, cuts processed until now is 145
+2022-03-21 15:57:57,432 INFO [decode.py:350] batch 12/?, cuts processed until now is 152
+2022-03-21 15:58:17,428 INFO [decode.py:350] batch 14/?, cuts processed until now is 165
+2022-03-21 15:58:39,131 INFO [decode.py:350] batch 16/?, cuts processed until now is 185
+2022-03-21 15:59:00,893 INFO [decode.py:350] batch 18/?, cuts processed until now is 205
+2022-03-21 15:59:21,853 INFO [decode.py:350] batch 20/?, cuts processed until now is 219
+2022-03-21 15:59:44,005 INFO [decode.py:350] batch 22/?, cuts processed until now is 262
+2022-03-21 16:00:05,041 INFO [decode.py:350] batch 24/?, cuts processed until now is 281
+2022-03-21 16:00:28,148 INFO [decode.py:350] batch 26/?, cuts processed until now is 297
+2022-03-21 16:00:49,235 INFO [decode.py:350] batch 28/?, cuts processed until now is 316
+2022-03-21 16:01:11,127 INFO [decode.py:350] batch 30/?, cuts processed until now is 334
+2022-03-21 16:01:33,292 INFO [decode.py:350] batch 32/?, cuts processed until now is 358
+2022-03-21 16:01:54,621 INFO [decode.py:350] batch 34/?, cuts processed until now is 389
+2022-03-21 16:02:15,415 INFO [decode.py:350] batch 36/?, cuts processed until now is 408
+2022-03-21 16:02:35,998 INFO [decode.py:350] batch 38/?, cuts processed until now is 428
+2022-03-21 16:02:58,046 INFO [decode.py:350] batch 40/?, cuts processed until now is 441
+2022-03-21 16:03:20,083 INFO [decode.py:350] batch 42/?, cuts processed until now is 489
+2022-03-21 16:03:42,567 INFO [decode.py:350] batch 44/?, cuts processed until now is 560
+2022-03-21 16:04:03,063 INFO [decode.py:350] batch 46/?, cuts processed until now is 573
+2022-03-21 16:04:25,344 INFO [decode.py:350] batch 48/?, cuts processed until now is 589
+2022-03-21 16:04:48,266 INFO [decode.py:350] batch 50/?, cuts processed until now is 605
+2022-03-21 16:05:09,812 INFO [decode.py:350] batch 52/?, cuts processed until now is 622
+2022-03-21 16:05:32,201 INFO [decode.py:350] batch 54/?, cuts processed until now is 645
+2022-03-21 16:05:53,067 INFO [decode.py:350] batch 56/?, cuts processed until now is 672
+2022-03-21 16:06:15,562 INFO [decode.py:350] batch 58/?, cuts processed until now is 692
+2022-03-21 16:06:37,993 INFO [decode.py:350] batch 60/?, cuts processed until now is 729
+2022-03-21 16:06:59,468 INFO [decode.py:350] batch 62/?, cuts processed until now is 749
+2022-03-21 16:07:20,240 INFO [decode.py:350] batch 64/?, cuts processed until now is 761
+2022-03-21 16:07:41,764 INFO [decode.py:350] batch 66/?, cuts processed until now is 784
+2022-03-21 16:07:54,918 INFO [decode.py:350] batch 68/?, cuts processed until now is 807
+2022-03-21 16:08:16,409 INFO [decode.py:350] batch 70/?, cuts processed until now is 829
+2022-03-21 16:08:36,930 INFO [decode.py:350] batch 72/?, cuts processed until now is 858
+2022-03-21 16:08:52,990 INFO [decode.py:350] batch 74/?, cuts processed until now is 883
+2022-03-21 16:09:15,343 INFO [decode.py:350] batch 76/?, cuts processed until now is 898
+2022-03-21 16:09:37,183 INFO [decode.py:350] batch 78/?, cuts processed until now is 928
+2022-03-21 16:09:57,897 INFO [decode.py:350] batch 80/?, cuts processed until now is 949
+2022-03-21 16:10:19,055 INFO [decode.py:350] batch 82/?, cuts processed until now is 964
+2022-03-21 16:14:48,880 INFO [decode.py:350] batch 84/?, cuts processed until now is 979
+2022-03-21 16:24:09,638 INFO [decode.py:350] batch 86/?, cuts processed until now is 998
+2022-03-21 16:30:52,650 INFO [decode.py:350] batch 88/?, cuts processed until now is 1017
+2022-03-21 16:32:12,892 INFO [decode.py:350] batch 90/?, cuts processed until now is 1031
+2022-03-21 16:32:36,186 INFO [decode.py:350] batch 92/?, cuts processed until now is 1055
+2022-03-21 16:32:55,654 INFO [decode.py:350] batch 94/?, cuts processed until now is 1089
+2022-03-21 16:33:09,893 INFO [decode.py:350] batch 96/?, cuts processed until now is 1107
+2022-03-21 16:33:21,010 INFO [decode.py:350] batch 98/?, cuts processed until now is 1117
+2022-03-21 16:33:42,203 INFO [decode.py:350] batch 100/?, cuts processed until now is 1133
+2022-03-21 16:33:55,144 INFO [decode.py:350] batch 102/?, cuts processed until now is 1145
+2022-03-21 16:34:10,494 INFO [decode.py:350] batch 104/?, cuts processed until now is 1155
+2022-03-21 16:34:10,613 INFO [decode.py:367] The transcripts are stored in transducer_stateless/exp/beam_search/recogs-test-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 16:34:10,649 INFO [utils.py:406] [test-beam_4] %WER 6.36% [1809 / 28430, 265 ins, 609 del, 935 sub ]
+2022-03-21 16:34:10,746 INFO [decode.py:380] Wrote detailed error stats to transducer_stateless/exp/beam_search/errs-test-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 16:34:10,747 INFO [decode.py:397] 
+For test, WER of different settings are:
+beam_4	6.36	best for test
+
+2022-03-21 16:34:10,747 INFO [decode.py:489] Done!

log/beam_search/wer-summary-dev-beam_4-epoch-29-avg-11-beam-4.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_4	7.02

log/beam_search/wer-summary-test-beam_4-epoch-29-avg-11-beam-4.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_4	6.36

log/greedy_search/log-decode-epoch-29-avg-11-context-2-max-sym-per-frame-3-2022-03-21-14-41-55

@@ -0,0 +1,28 @@
+2022-03-21 14:41:55,508 INFO [decode.py:425] Decoding started
+2022-03-21 14:41:55,561 INFO [decode.py:431] Device: cuda:0
+2022-03-21 14:41:55,569 INFO [decode.py:441] {'feature_dim': 80, 'encoder_out_dim': 512, 'subsampling_factor': 4, 'attention_dim': 512, 'nhead': 8, 'dim_feedforward': 2048, 'num_encoder_layers': 12, 'vgg_frontend': False, 'epoch': 29, 'avg': 11, 'exp_dir': PosixPath('transducer_stateless/exp'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'decoding_method': 'greedy_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'res_dir': PosixPath('transducer_stateless/exp/greedy_search'), 'suffix': 'epoch-29-avg-11-context-2-max-sym-per-frame-3', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
+2022-03-21 14:41:55,570 INFO [decode.py:443] About to create model
+2022-03-21 14:41:56,509 INFO [decode.py:454] averaging ['transducer_stateless/exp/epoch-19.pt', 'transducer_stateless/exp/epoch-20.pt', 'transducer_stateless/exp/epoch-21.pt', 'transducer_stateless/exp/epoch-22.pt', 'transducer_stateless/exp/epoch-23.pt', 'transducer_stateless/exp/epoch-24.pt', 'transducer_stateless/exp/epoch-25.pt', 'transducer_stateless/exp/epoch-26.pt', 'transducer_stateless/exp/epoch-27.pt', 'transducer_stateless/exp/epoch-28.pt', 'transducer_stateless/exp/epoch-29.pt']
+2022-03-21 14:43:16,454 INFO [decode.py:463] Number of model parameters: 84007924
+2022-03-21 14:43:16,455 INFO [asr_datamodule.py:357] About to get dev cuts
+2022-03-21 14:43:16,492 INFO [asr_datamodule.py:362] About to get test cuts
+2022-03-21 14:43:16,563 INFO [asr_datamodule.py:300] About to create dev dataset
+2022-03-21 14:43:16,564 INFO [asr_datamodule.py:319] About to create dev dataloader
+2022-03-21 14:43:18,487 INFO [decode.py:350] batch 0/?, cuts processed until now is 10
+2022-03-21 14:44:22,097 INFO [decode.py:367] The transcripts are stored in transducer_stateless/exp/greedy_search/recogs-dev-greedy_search-epoch-29-avg-11-context-2-max-sym-per-frame-3.txt
+2022-03-21 14:44:22,130 INFO [utils.py:406] [dev-greedy_search] %WER 7.19% [1311 / 18226, 192 ins, 434 del, 685 sub ]
+2022-03-21 14:44:22,212 INFO [decode.py:380] Wrote detailed error stats to transducer_stateless/exp/greedy_search/errs-dev-greedy_search-epoch-29-avg-11-context-2-max-sym-per-frame-3.txt
+2022-03-21 14:44:22,212 INFO [decode.py:397] 
+For dev, WER of different settings are:
+greedy_search	7.19	best for dev
+
+2022-03-21 14:44:23,386 INFO [decode.py:350] batch 0/?, cuts processed until now is 14
+2022-03-21 14:46:31,785 INFO [decode.py:350] batch 100/?, cuts processed until now is 1133
+2022-03-21 14:46:35,935 INFO [decode.py:367] The transcripts are stored in transducer_stateless/exp/greedy_search/recogs-test-greedy_search-epoch-29-avg-11-context-2-max-sym-per-frame-3.txt
+2022-03-21 14:46:35,971 INFO [utils.py:406] [test-greedy_search] %WER 6.70% [1906 / 28430, 224 ins, 745 del, 937 sub ]
+2022-03-21 14:46:36,082 INFO [decode.py:380] Wrote detailed error stats to transducer_stateless/exp/greedy_search/errs-test-greedy_search-epoch-29-avg-11-context-2-max-sym-per-frame-3.txt
+2022-03-21 14:46:36,083 INFO [decode.py:397] 
+For test, WER of different settings are:
+greedy_search	6.7	best for test
+
+2022-03-21 14:46:36,083 INFO [decode.py:489] Done!

log/greedy_search/wer-summary-dev-greedy_search-epoch-29-avg-11-context-2-max-sym-per-frame-3.txt

@@ -0,0 +1,2 @@
+settings	WER
+greedy_search	7.19

log/greedy_search/wer-summary-test-greedy_search-epoch-29-avg-11-context-2-max-sym-per-frame-3.txt

@@ -0,0 +1,2 @@
+settings	WER
+greedy_search	6.7

log/modified_beam_search/log-decode-epoch-29-avg-11-beam-4-2022-03-21-15-36-23

@@ -0,0 +1,5 @@
+2022-03-21 15:36:23,609 INFO [decode.py:425] Decoding started
+2022-03-21 15:36:23,667 INFO [decode.py:431] Device: cuda:0
+2022-03-21 15:36:23,674 INFO [decode.py:441] {'feature_dim': 80, 'encoder_out_dim': 512, 'subsampling_factor': 4, 'attention_dim': 512, 'nhead': 8, 'dim_feedforward': 2048, 'num_encoder_layers': 12, 'vgg_frontend': False, 'epoch': 29, 'avg': 11, 'exp_dir': PosixPath('transducer_stateless/exp'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'decoding_method': 'modified_beam_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'res_dir': PosixPath('transducer_stateless/exp/modified_beam_search'), 'suffix': 'epoch-29-avg-11-beam-4', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
+2022-03-21 15:36:23,674 INFO [decode.py:443] About to create model
+2022-03-21 15:36:24,366 INFO [decode.py:454] averaging ['transducer_stateless/exp/epoch-19.pt', 'transducer_stateless/exp/epoch-20.pt', 'transducer_stateless/exp/epoch-21.pt', 'transducer_stateless/exp/epoch-22.pt', 'transducer_stateless/exp/epoch-23.pt', 'transducer_stateless/exp/epoch-24.pt', 'transducer_stateless/exp/epoch-25.pt', 'transducer_stateless/exp/epoch-26.pt', 'transducer_stateless/exp/epoch-27.pt', 'transducer_stateless/exp/epoch-28.pt', 'transducer_stateless/exp/epoch-29.pt']

log/modified_beam_search/log-decode-epoch-29-avg-11-beam-4-2022-03-21-15-36-41

@@ -0,0 +1,112 @@
+2022-03-21 15:36:41,949 INFO [decode.py:425] Decoding started
+2022-03-21 15:36:42,033 INFO [decode.py:431] Device: cuda:0
+2022-03-21 15:36:42,036 INFO [decode.py:441] {'feature_dim': 80, 'encoder_out_dim': 512, 'subsampling_factor': 4, 'attention_dim': 512, 'nhead': 8, 'dim_feedforward': 2048, 'num_encoder_layers': 12, 'vgg_frontend': False, 'epoch': 29, 'avg': 11, 'exp_dir': PosixPath('transducer_stateless/exp'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'decoding_method': 'modified_beam_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'res_dir': PosixPath('transducer_stateless/exp/modified_beam_search'), 'suffix': 'epoch-29-avg-11-beam-4', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
+2022-03-21 15:36:42,036 INFO [decode.py:443] About to create model
+2022-03-21 15:36:42,641 INFO [decode.py:454] averaging ['transducer_stateless/exp/epoch-19.pt', 'transducer_stateless/exp/epoch-20.pt', 'transducer_stateless/exp/epoch-21.pt', 'transducer_stateless/exp/epoch-22.pt', 'transducer_stateless/exp/epoch-23.pt', 'transducer_stateless/exp/epoch-24.pt', 'transducer_stateless/exp/epoch-25.pt', 'transducer_stateless/exp/epoch-26.pt', 'transducer_stateless/exp/epoch-27.pt', 'transducer_stateless/exp/epoch-28.pt', 'transducer_stateless/exp/epoch-29.pt']
+2022-03-21 15:37:55,175 INFO [decode.py:463] Number of model parameters: 84007924
+2022-03-21 15:37:55,175 INFO [asr_datamodule.py:357] About to get dev cuts
+2022-03-21 15:37:55,210 INFO [asr_datamodule.py:362] About to get test cuts
+2022-03-21 15:37:55,276 INFO [asr_datamodule.py:300] About to create dev dataset
+2022-03-21 15:37:55,277 INFO [asr_datamodule.py:319] About to create dev dataloader
+2022-03-21 15:37:57,065 INFO [decode.py:350] batch 0/?, cuts processed until now is 10
+2022-03-21 15:37:59,913 INFO [decode.py:350] batch 2/?, cuts processed until now is 33
+2022-03-21 15:38:02,939 INFO [decode.py:350] batch 4/?, cuts processed until now is 45
+2022-03-21 15:38:05,841 INFO [decode.py:350] batch 6/?, cuts processed until now is 67
+2022-03-21 15:38:08,783 INFO [decode.py:350] batch 8/?, cuts processed until now is 77
+2022-03-21 15:38:11,617 INFO [decode.py:350] batch 10/?, cuts processed until now is 96
+2022-03-21 15:38:14,504 INFO [decode.py:350] batch 12/?, cuts processed until now is 101
+2022-03-21 15:38:17,404 INFO [decode.py:350] batch 14/?, cuts processed until now is 111
+2022-03-21 15:38:20,264 INFO [decode.py:350] batch 16/?, cuts processed until now is 125
+2022-03-21 15:38:22,983 INFO [decode.py:350] batch 18/?, cuts processed until now is 140
+2022-03-21 15:38:25,709 INFO [decode.py:350] batch 20/?, cuts processed until now is 158
+2022-03-21 15:38:28,478 INFO [decode.py:350] batch 22/?, cuts processed until now is 184
+2022-03-21 15:38:31,165 INFO [decode.py:350] batch 24/?, cuts processed until now is 198
+2022-03-21 15:38:34,191 INFO [decode.py:350] batch 26/?, cuts processed until now is 209
+2022-03-21 15:38:36,809 INFO [decode.py:350] batch 28/?, cuts processed until now is 223
+2022-03-21 15:38:39,821 INFO [decode.py:350] batch 30/?, cuts processed until now is 237
+2022-03-21 15:38:42,888 INFO [decode.py:350] batch 32/?, cuts processed until now is 256
+2022-03-21 15:38:45,766 INFO [decode.py:350] batch 34/?, cuts processed until now is 278
+2022-03-21 15:38:48,813 INFO [decode.py:350] batch 36/?, cuts processed until now is 294
+2022-03-21 15:38:51,730 INFO [decode.py:350] batch 38/?, cuts processed until now is 306
+2022-03-21 15:38:54,854 INFO [decode.py:350] batch 40/?, cuts processed until now is 314
+2022-03-21 15:38:57,758 INFO [decode.py:350] batch 42/?, cuts processed until now is 345
+2022-03-21 15:38:59,834 INFO [decode.py:350] batch 44/?, cuts processed until now is 378
+2022-03-21 15:39:02,701 INFO [decode.py:350] batch 46/?, cuts processed until now is 388
+2022-03-21 15:39:05,385 INFO [decode.py:350] batch 48/?, cuts processed until now is 404
+2022-03-21 15:39:07,430 INFO [decode.py:350] batch 50/?, cuts processed until now is 412
+2022-03-21 15:39:10,290 INFO [decode.py:350] batch 52/?, cuts processed until now is 425
+2022-03-21 15:39:12,050 INFO [decode.py:350] batch 54/?, cuts processed until now is 435
+2022-03-21 15:39:14,847 INFO [decode.py:350] batch 56/?, cuts processed until now is 446
+2022-03-21 15:39:16,517 INFO [decode.py:350] batch 58/?, cuts processed until now is 458
+2022-03-21 15:39:19,794 INFO [decode.py:350] batch 60/?, cuts processed until now is 474
+2022-03-21 15:39:22,030 INFO [decode.py:350] batch 62/?, cuts processed until now is 483
+2022-03-21 15:39:23,700 INFO [decode.py:350] batch 64/?, cuts processed until now is 493
+2022-03-21 15:39:26,665 INFO [decode.py:350] batch 66/?, cuts processed until now is 507
+2022-03-21 15:39:26,786 INFO [decode.py:367] The transcripts are stored in transducer_stateless/exp/modified_beam_search/recogs-dev-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 15:39:26,812 INFO [utils.py:406] [dev-beam_4] %WER 6.91% [1259 / 18226, 192 ins, 394 del, 673 sub ]
+2022-03-21 15:39:26,873 INFO [decode.py:380] Wrote detailed error stats to transducer_stateless/exp/modified_beam_search/errs-dev-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 15:39:26,874 INFO [decode.py:397] 
+For dev, WER of different settings are:
+beam_4	6.91	best for dev
+
+2022-03-21 15:39:28,685 INFO [decode.py:350] batch 0/?, cuts processed until now is 14
+2022-03-21 15:39:31,726 INFO [decode.py:350] batch 2/?, cuts processed until now is 51
+2022-03-21 15:39:34,578 INFO [decode.py:350] batch 4/?, cuts processed until now is 67
+2022-03-21 15:39:37,519 INFO [decode.py:350] batch 6/?, cuts processed until now is 104
+2022-03-21 15:39:40,418 INFO [decode.py:350] batch 8/?, cuts processed until now is 118
+2022-03-21 15:39:43,196 INFO [decode.py:350] batch 10/?, cuts processed until now is 145
+2022-03-21 15:39:46,351 INFO [decode.py:350] batch 12/?, cuts processed until now is 152
+2022-03-21 15:39:49,219 INFO [decode.py:350] batch 14/?, cuts processed until now is 165
+2022-03-21 15:39:52,253 INFO [decode.py:350] batch 16/?, cuts processed until now is 185
+2022-03-21 15:39:55,087 INFO [decode.py:350] batch 18/?, cuts processed until now is 205
+2022-03-21 15:39:57,945 INFO [decode.py:350] batch 20/?, cuts processed until now is 219
+2022-03-21 15:40:00,755 INFO [decode.py:350] batch 22/?, cuts processed until now is 262
+2022-03-21 15:40:03,577 INFO [decode.py:350] batch 24/?, cuts processed until now is 281
+2022-03-21 15:40:06,747 INFO [decode.py:350] batch 26/?, cuts processed until now is 297
+2022-03-21 15:40:09,632 INFO [decode.py:350] batch 28/?, cuts processed until now is 316
+2022-03-21 15:40:12,595 INFO [decode.py:350] batch 30/?, cuts processed until now is 334
+2022-03-21 15:40:15,515 INFO [decode.py:350] batch 32/?, cuts processed until now is 358
+2022-03-21 15:40:18,285 INFO [decode.py:350] batch 34/?, cuts processed until now is 389
+2022-03-21 15:40:21,168 INFO [decode.py:350] batch 36/?, cuts processed until now is 408
+2022-03-21 15:40:23,897 INFO [decode.py:350] batch 38/?, cuts processed until now is 428
+2022-03-21 15:40:26,703 INFO [decode.py:350] batch 40/?, cuts processed until now is 441
+2022-03-21 15:40:29,504 INFO [decode.py:350] batch 42/?, cuts processed until now is 489
+2022-03-21 15:40:32,393 INFO [decode.py:350] batch 44/?, cuts processed until now is 560
+2022-03-21 15:40:35,057 INFO [decode.py:350] batch 46/?, cuts processed until now is 573
+2022-03-21 15:40:38,063 INFO [decode.py:350] batch 48/?, cuts processed until now is 589
+2022-03-21 15:40:41,215 INFO [decode.py:350] batch 50/?, cuts processed until now is 605
+2022-03-21 15:40:44,171 INFO [decode.py:350] batch 52/?, cuts processed until now is 622
+2022-03-21 15:40:47,088 INFO [decode.py:350] batch 54/?, cuts processed until now is 645
+2022-03-21 15:40:49,917 INFO [decode.py:350] batch 56/?, cuts processed until now is 672
+2022-03-21 15:40:52,974 INFO [decode.py:350] batch 58/?, cuts processed until now is 692
+2022-03-21 15:40:55,807 INFO [decode.py:350] batch 60/?, cuts processed until now is 729
+2022-03-21 15:40:58,701 INFO [decode.py:350] batch 62/?, cuts processed until now is 749
+2022-03-21 15:41:01,465 INFO [decode.py:350] batch 64/?, cuts processed until now is 761
+2022-03-21 15:41:04,269 INFO [decode.py:350] batch 66/?, cuts processed until now is 784
+2022-03-21 15:41:05,969 INFO [decode.py:350] batch 68/?, cuts processed until now is 807
+2022-03-21 15:41:08,783 INFO [decode.py:350] batch 70/?, cuts processed until now is 829
+2022-03-21 15:41:11,590 INFO [decode.py:350] batch 72/?, cuts processed until now is 858
+2022-03-21 15:41:13,685 INFO [decode.py:350] batch 74/?, cuts processed until now is 883
+2022-03-21 15:41:16,555 INFO [decode.py:350] batch 76/?, cuts processed until now is 898
+2022-03-21 15:41:19,318 INFO [decode.py:350] batch 78/?, cuts processed until now is 928
+2022-03-21 15:41:22,092 INFO [decode.py:350] batch 80/?, cuts processed until now is 949
+2022-03-21 15:41:24,954 INFO [decode.py:350] batch 82/?, cuts processed until now is 964
+2022-03-21 15:41:27,959 INFO [decode.py:350] batch 84/?, cuts processed until now is 979
+2022-03-21 15:41:30,849 INFO [decode.py:350] batch 86/?, cuts processed until now is 998
+2022-03-21 15:41:33,047 INFO [decode.py:350] batch 88/?, cuts processed until now is 1017
+2022-03-21 15:41:35,357 INFO [decode.py:350] batch 90/?, cuts processed until now is 1031
+2022-03-21 15:41:38,318 INFO [decode.py:350] batch 92/?, cuts processed until now is 1055
+2022-03-21 15:41:41,009 INFO [decode.py:350] batch 94/?, cuts processed until now is 1089
+2022-03-21 15:41:43,058 INFO [decode.py:350] batch 96/?, cuts processed until now is 1107
+2022-03-21 15:41:44,573 INFO [decode.py:350] batch 98/?, cuts processed until now is 1117
+2022-03-21 15:41:47,809 INFO [decode.py:350] batch 100/?, cuts processed until now is 1133
+2022-03-21 15:41:49,516 INFO [decode.py:350] batch 102/?, cuts processed until now is 1145
+2022-03-21 15:41:51,747 INFO [decode.py:350] batch 104/?, cuts processed until now is 1155
+2022-03-21 15:41:51,865 INFO [decode.py:367] The transcripts are stored in transducer_stateless/exp/modified_beam_search/recogs-test-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 15:41:51,899 INFO [utils.py:406] [test-beam_4] %WER 6.33% [1799 / 28430, 230 ins, 639 del, 930 sub ]
+2022-03-21 15:41:51,990 INFO [decode.py:380] Wrote detailed error stats to transducer_stateless/exp/modified_beam_search/errs-test-beam_4-epoch-29-avg-11-beam-4.txt
+2022-03-21 15:41:51,990 INFO [decode.py:397] 
+For test, WER of different settings are:
+beam_4	6.33	best for test
+
+2022-03-21 15:41:51,990 INFO [decode.py:489] Done!

log/modified_beam_search/wer-summary-dev-beam_4-epoch-29-avg-11-beam-4.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_4	6.91

log/modified_beam_search/wer-summary-test-beam_4-epoch-29-avg-11-beam-4.txt

@@ -0,0 +1,2 @@
+settings	WER
+beam_4	6.33

test_wavs/RESULTS.md

@@ -1,28 +1,27 @@
 You can use the following command to test the pretrained.py and the pretrained files:
 ```
-CUDA_VISIBLE_DEVICES='1' python transducer_stateless/pretrained.py --checkpoint icefall_asr_tedlium3_transducer_stateless/exp/pretrained_average_14_to_29.pt --bpe-model icefall_asr_tedlium3_transducer_stateless/data/lang_bpe_500/bpe.model --method greedy_search icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W01.wav icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W02.wav icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W03.wav
+CUDA_VISIBLE_DEVICES='1' python transducer_stateless/pretrained.py --checkpoint icefall_asr_tedlium3_transducer_stateless/exp/pretrained_average_19_to_29.pt --bpe-model icefall_asr_tedlium3_transducer_stateless/data/lang_bpe_500/bpe.model --method greedy_search icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W01.wav icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W02.wav icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W03.wav
 ```
 
 The running results are as follows:
 ```
-2022-03-03 14:59:21,847 INFO [pretrained.py:251] {'sample_rate': 16000, 'feature_dim': 80, 'encoder_out_dim': 512, 'subsampling_factor': 4, 'attention_dim': 512, 'nhead': 8, 'dim_feedforward': 2048, 'num_encoder_layers': 12, 'vgg_frontend': False, 'checkpoint': 'icefall_asr_tedlium3_transducer_stateless/exp/pretrained_average_14_to_29.pt', 'bpe_model': 'icefall_asr_tedlium3_transducer_stateless/data/lang_bpe_500/bpe.model', 'method': 'greedy_search', 'sound_files': ['icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W01.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W02.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W03.wav'], 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
-2022-03-03 14:59:21,920 INFO [pretrained.py:257] device: cuda:0
-2022-03-03 14:59:21,920 INFO [pretrained.py:259] Creating model
-2022-03-03 14:59:30,749 INFO [pretrained.py:268] Constructing Fbank computer
-2022-03-03 14:59:30,750 INFO [pretrained.py:278] Reading sound files: ['icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W01.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W02.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W03.wav']
-2022-03-03 14:59:30,768 INFO [pretrained.py:284] Decoding started
-2022-03-03 14:59:30,908 INFO [pretrained.py:304] Using greedy_search
-2022-03-03 14:59:31,757 INFO [pretrained.py:332]
+2022-03-21 16:01:53,281 INFO [pretrained.py:251] {'sample_rate': 16000, 'feature_dim': 80, 'encoder_out_dim': 512, 'subsampling_factor': 4, 'attention_dim': 512, 'nhead': 8, 'dim_feedforward': 2048, 'num_encoder_layers': 12, 'vgg_frontend': False, 'env_info': {'k2-version': '1.13', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '5ee082ea55f50e8bd42203ba266945ea5a236ab8', 'k2-git-date': 'Sun Feb 27 09:00:48 2022', 'lhotse-version': '1.0.0.dev+git.d917411.clean', 'torch-cuda-available': True, 'torch-cuda-version': '10.1', 'python-version': '3.8', 'icefall-git-branch': 'tedlium3-pruned-transducer-stateless-recipe', 'icefall-git-sha1': 'ad28c8c-dirty', 'icefall-git-date': 'Fri Mar 18 11:39:06 2022', 'icefall-path': '/ceph-meixu/luomingshuang/icefall', 'k2-path': '/ceph-meixu/luomingshuang/k2/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-meixu/luomingshuang/anaconda3/envs/k2-python/lib/python3.8/site-packages/lhotse-1.0.0.dev0+git.d917411.clean-py3.8.egg/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-2-0307200233-b554c565c-lf9qd', 'IP address': '10.177.74.201'}, 'checkpoint': 'icefall_asr_tedlium3_transducer_stateless/exp/pretrained_average_19_to_29.pt', 'bpe_model': 'icefall_asr_tedlium3_transducer_stateless/data/lang_bpe_500/bpe.model', 'method': 'greedy_search', 'sound_files': ['icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W01.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W02.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W03.wav'], 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
+2022-03-21 16:01:53,282 INFO [pretrained.py:257] device: cuda:0
+2022-03-21 16:01:53,282 INFO [pretrained.py:259] Creating model
+2022-03-21 16:02:09,338 INFO [pretrained.py:268] Constructing Fbank computer
+2022-03-21 16:02:09,341 INFO [pretrained.py:278] Reading sound files: ['icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W01.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W02.wav', 'icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W03.wav']
+2022-03-21 16:02:09,374 INFO [pretrained.py:284] Decoding started
+2022-03-21 16:02:09,526 INFO [pretrained.py:304] Using greedy_search
+2022-03-21 16:02:10,931 INFO [pretrained.py:332]
 icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W01.wav:
-isn 't it i don 't live there but i did journey on a twenty seven thousand mile trip for two years to the fastest growing and widest counties in america what is a whitopia i define whitopia in three ways first whitopia has posted at least six percent
+choice isn 't it i don 't live there but i did journey on a twenty seven thousand mile trip for two years to the fastest scroing and whiteest counties in america what is a whitopia i define whitopia in three ways first whitopia has posted at least six percent
 
 icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W02.wav:
-population growth since two thousand secondly the majority of that growth comes from white migrants and third the whitopia has an ineffable charm a pleasant look and feel a junast qua to learn how and why whitopias are ticking i immersed myself for several months apiece in three of them first
+population growth since two thousand secondly the majority of that growth comes from white migrants and third the whitopia has an ineffable charm a pleasant look and feel a genose clas to learn how and why whitopias are ticking i immerse myself for several months apiece in three of them first
 
 icefall_asr_tedlium3_transducer_stateless/test_wavs/RichBenjamin_2015W03.wav:
-st george utah second queurdalen idaho and third for sie county georgia first stop st george a beautiful town of red rock landscapes in the one thousand eight hundred and fifty 's brigham young dispatched families to st george to grow cotton because of the hot arid climate and so they called it utah 's dixie and the name sticks to this day
+st george utah second queur d 'alne idaho and third forcie county georgia first stop st george a beautiful town of red rock landscapes in the one thousand eight hundred and fifty 's brigham young dispatched families to st george to grow cotton because of the hot arid climate and so they called it utah 's dixie and the name sticks to this day
 
 
-2022-03-03 14:59:31,757 INFO [pretrained.py:334] Decoding Done
-
+2022-03-21 16:02:10,931 INFO [pretrained.py:334] Decoding Done
 ```