Hugging Face's logo

csukuangfj
/
icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21

Model card Files Community
csukuangfj commited on
Commit
d3c37c8
•
1 Parent(s): ba27a9b

First commit.

Browse files
Files changed (27) hide show
  1. README.md +154 -0
  2. data/lang_bpe_500/bpe.model +3 -0
  3. exp/pretrained.pt +3 -0
  4. log/errs-test-clean-beam_4-epoch-57-avg-17-beam-4.txt +0 -0
  5. log/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt +0 -0
  6. log/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt +0 -0
  7. log/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt +0 -0
  8. log/errs-test-other-beam_4-epoch-57-avg-17-beam-4.txt +0 -0
  9. log/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt +0 -0
  10. log/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt +0 -0
  11. log/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt +0 -0
  12. log/log-decode-epoch-57-avg-17-beam-4-2022-02-20-11-57-47 +233 -0
  13. log/log-decode-epoch-57-avg-17-context-2-max-sym-per-frame-1-2022-02-20-11-27-46 +29 -0
  14. log/log-decode-epoch-57-avg-17-context-2-max-sym-per-frame-2-2022-02-21-11-31-01 +29 -0
  15. log/log-decode-epoch-57-avg-17-context-2-max-sym-per-frame-3-2022-02-21-12-13-11 +29 -0
  16. log/recogs-test-clean-beam_4-epoch-57-avg-17-beam-4.txt +0 -0
  17. log/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt +0 -0
  18. log/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt +0 -0
  19. log/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt +0 -0
  20. log/recogs-test-other-beam_4-epoch-57-avg-17-beam-4.txt +0 -0
  21. log/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt +0 -0
  22. log/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt +0 -0
  23. log/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt +0 -0
  24. test_wavs/1089-134686-0001.wav +0 -0
  25. test_wavs/1221-135766-0001.wav +0 -0
  26. test_wavs/1221-135766-0002.wav +0 -0
  27. test_wavs/trans.txt +3 -0
README.md ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Introduction
2
+
3
+ This repo contains pre-trained model using
4
+ <https://github.com/k2-fsa/icefall/pull/213>.
5
+
6
+ It is trained on train-clean-100 subset of the LibriSpeech dataset.
7
+ Also, it uses the `S` subset from GigaSpeech as extra training data.
8
+
9
+ ## How to clone this repo
10
+ ```
11
+ sudo apt-get install git-lfs
12
+ git clone https://huggingface.co/csukuangfj/icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21
13
+
14
+ cd icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21
15
+ git lfs pull
16
+ ```
17
+
18
+ **Catuion**: You have to run `git lfs pull`. Otherwise, you will be SAD later.
19
+
20
+ The model in this repo is trained using the commit `TODO`.
21
+
22
+ You can use
23
+
24
+ ```
25
+ git clone https://github.com/k2-fsa/icefall
26
+ cd icefall
27
+ git checkout TODO
28
+ ```
29
+ to download `icefall`.
30
+
31
+ You can find the model information by visiting <https://github.com/k2-fsa/icefall/blob/TODO/egs/librispeech/ASR/transducer_stateless_multi_datasets/train.py#L198>.
32
+
33
+ In short, the encoder is a Conformer model with 8 heads, 12 encoder layers, 512-dim attention, 2048-dim feedforward;
34
+ the decoder contains a 1024-dim embedding layer and a Conv1d with kernel size 2.
35
+
36
+ The decoder architecture is modified from
37
+ [Rnn-Transducer with Stateless Prediction Network](https://ieeexplore.ieee.org/document/9054419).
38
+ A Conv1d layer is placed right after the input embedding layer.
39
+
40
+ -----
41
+
42
+ ## Description
43
+
44
+ This repo provides pre-trained transducer Conformer model for the LibriSpeech dataset
45
+ using [icefall][icefall]. There are no RNNs in the decoder. The decoder is stateless
46
+ and contains only an embedding layer and a Conv1d.
47
+
48
+ The commands for training are:
49
+
50
+ ```
51
+ cd egs/librispeech/ASR/
52
+ ./prepare.sh
53
+ ./prepare_giga_speech.sh
54
+
55
+ export CUDA_VISIBLE_DEVICES="0,1"
56
+
57
+ ./transducer_stateless_multi_datasets/train.py \
58
+ --world-size 2 \
59
+ --num-epochs 60 \
60
+ --start-epoch 0 \
61
+ --exp-dir transducer_stateless_multi_datasets/exp-100-2 \
62
+ --full-libri 0 \
63
+ --max-duration 300 \
64
+ --lr-factor 1 \
65
+ --bpe-model data/lang_bpe_500/bpe.model \
66
+ --modified-transducer-prob 0.25
67
+ --giga-prob 0.2
68
+ ```
69
+
70
+ The tensorboard training log can be found at
71
+ <https://tensorboard.dev/experiment/qUEKzMnrTZmOz1EXPda9RA/>
72
+
73
+ The command for decoding is:
74
+ ```
75
+ epoch=57
76
+ avg=17
77
+
78
+ ## greedy search
79
+ for epoch in 57; do
80
+ for avg in 17; do
81
+ for sym in 1 2 3; do
82
+ ./transducer_stateless_multi_datasets/decode.py \
83
+ --epoch $epoch \
84
+ --avg $avg \
85
+ --exp-dir transducer_stateless_multi_datasets/exp-100-2 \
86
+ --bpe-model ./data/lang_bpe_500/bpe.model \
87
+ --max-duration 100 \
88
+ --context-size 2 \
89
+ --max-sym-per-frame $sym
90
+ done
91
+ done
92
+ done
93
+
94
+ ## modified beam search
95
+
96
+ epoch=57
97
+ avg=17
98
+ ./transducer_stateless_multi_datasets/decode.py \
99
+ --epoch $epoch \
100
+ --avg $avg \
101
+ --exp-dir transducer_stateless_multi_datasets/exp-100-2 \
102
+ --bpe-model ./data/lang_bpe_500/bpe.model \
103
+ --max-duration 100 \
104
+ --context-size 2 \
105
+ --decoding-method modified_beam_search \
106
+ --beam-size 4
107
+ ```
108
+
109
+ You can find the decoding log for the above command in this
110
+ repo (in the folder `log`).
111
+
112
+ The WERs for the test datasets are
113
+
114
+ | | test-clean | test-other | comment |
115
+ |-------------------------------------|------------|------------|------------------------------------------|
116
+ | greedy search (max sym per frame 1) | 6.34 | 16.7 | --epoch 57, --avg 17, --max-duration 100 |
117
+ | greedy search (max sym per frame 2) | 6.34 | 16.7 | --epoch 57, --avg 17, --max-duration 100 |
118
+ | greedy search (max sym per frame 3) | 6.34 | 16.7 | --epoch 57, --avg 17, --max-duration 100 |
119
+ | modified beam search (beam size 4) | 6.31 | 16.3 | --epoch 57, --avg 17, --max-duration 100 |
120
+
121
+
122
+ # File description
123
+
124
+ - [log][log], this directory contains the decoding log and decoding results
125
+ - [test_wavs][test_wavs], this directory contains wave files for testing the pre-trained model
126
+ - [data][data], this directory contains files generated by [prepare.sh][prepare]
127
+ - [exp][exp], this directory contains only one file: `preprained.pt`
128
+
129
+ `exp/pretrained.pt` is generated by the following command:
130
+
131
+ ```bash
132
+ ./transducer_stateless_multi_datasets/export.py \
133
+ --epoch 57 \
134
+ --avg 17 \
135
+ --bpe-model data/lang_bpe_500/bpe.model \
136
+ --exp-dir transducer_stateless_multi_datasets/exp-full
137
+ ```
138
+
139
+ **HINT**: To use `pretrained.pt` to compute the WER for test-clean and test-other,
140
+ just do the following:
141
+ ```
142
+ cp icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21/exp/pretrained.pt \
143
+ /path/to/icefall/egs/librispeech/ASR/transducer_stateless_multi_datasets/exp/epoch-999.pt
144
+ ```
145
+ and pass `--epoch 999 --avg 1` to `transducer_stateless_multi_datasets/decode.py`.
146
+
147
+
148
+ [icefall]: https://github.com/k2-fsa/icefall
149
+ [prepare]: https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/prepare.sh
150
+ [exp]: https://huggingface.co/csukuangfj/icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21/tree/main/exp
151
+ [data]: https://huggingface.co/csukuangfj/icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21/tree/main/data
152
+ [test_wavs]: https://huggingface.co/csukuangfj/icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21/tree/main/test_wavs
153
+ [log]: https://huggingface.co/csukuangfj/icefall-asr-librispeech-100h-transducer-stateless-multi-datasets-bpe-500-2022-02-21/tree/main/log
154
+ [icefall]: https://github.com/k2-fsa/icefall
data/lang_bpe_500/bpe.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c53433de083c4a6ad12d034550ef22de68cec62c4f58932a7b6b8b2f1e743fa5
3
+ size 244865
exp/pretrained.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f67cf73b6a04e0bee38f736ce66e7cdb62060f20c0fd7d26a16f9693c7601e20
3
+ size 336185613
log/errs-test-clean-beam_4-epoch-57-avg-17-beam-4.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/errs-test-other-beam_4-epoch-57-avg-17-beam-4.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/log-decode-epoch-57-avg-17-beam-4-2022-02-20-11-57-47 ADDED
@@ -0,0 +1,233 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-02-20 11:57:47,740 INFO [decode.py:421] Decoding started
2
+ 2022-02-20 11:57:47,740 INFO [decode.py:427] Device: cpu
3
+ 2022-02-20 11:57:47,742 INFO [decode.py:436] {'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': 'f4fefe4882bc0ae59af951da3f47335d5495ef71', 'k2-git-date': 'Thu Feb 10 15:16:02 2022', 'lhotse-version': '1.0.0.dev+missing.version.file', 'torch-cuda-available': False, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'multiple-datasets', 'icefall-git-sha1': '61b0019-dirty', 'icefall-git-date': 'Thu Feb 17 18:34:48 2022', 'icefall-path': '/ceph-fj/fangjun/open-source-2/icefall-multi-datasets', 'k2-path': '/ceph-fj/fangjun/open-source-2/k2-multi-datasets/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-fj/fangjun/open-source-2/lhotse-multi-datasets/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-5-0816110343-9647676d5-sqn62', 'IP address': '10.177.63.199'}, 'epoch': 57, 'avg': 17, 'exp_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2'), 'bpe_model': './data/lang_bpe_500/bpe.model', 'decoding_method': 'modified_beam_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'manifest_dir': PosixPath('data/fbank'), 'on_the_fly_feats': False, 'res_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2/modified_beam_search'), 'suffix': 'epoch-57-avg-17-beam-4', 'blank_id': 0, 'vocab_size': 500}
4
+ 2022-02-20 11:57:47,743 INFO [decode.py:438] About to create model
5
+ 2022-02-20 11:57:48,209 INFO [decode.py:449] averaging ['transducer_stateless_multi_datasets/exp-100-2/epoch-41.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-42.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-43.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-44.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-45.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-46.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-47.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-48.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-49.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-50.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-51.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-52.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-53.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-54.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-55.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-56.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-57.pt']
6
+ 2022-02-20 11:58:06,598 INFO [decode.py:460] Number of model parameters: 84521448
7
+ 2022-02-20 11:58:06,598 INFO [librispeech.py:58] About to get test-clean cuts from data/fbank/cuts_test-clean.json.gz
8
+ 2022-02-20 11:58:06,782 INFO [librispeech.py:63] About to get test-other cuts from data/fbank/cuts_test-other.json.gz
9
+ 2022-02-20 11:58:18,257 INFO [decode.py:346] batch 0/?, cuts processed until now is 20
10
+ 2022-02-20 11:58:40,606 INFO [decode.py:346] batch 2/?, cuts processed until now is 64
11
+ 2022-02-20 11:59:02,607 INFO [decode.py:346] batch 4/?, cuts processed until now is 81
12
+ 2022-02-20 11:59:25,109 INFO [decode.py:346] batch 6/?, cuts processed until now is 121
13
+ 2022-02-20 11:59:46,773 INFO [decode.py:346] batch 8/?, cuts processed until now is 133
14
+ 2022-02-20 12:00:08,430 INFO [decode.py:346] batch 10/?, cuts processed until now is 168
15
+ 2022-02-20 12:00:30,120 INFO [decode.py:346] batch 12/?, cuts processed until now is 202
16
+ 2022-02-20 12:00:56,361 INFO [decode.py:346] batch 14/?, cuts processed until now is 217
17
+ 2022-02-20 12:01:19,746 INFO [decode.py:346] batch 16/?, cuts processed until now is 241
18
+ 2022-02-20 12:01:42,299 INFO [decode.py:346] batch 18/?, cuts processed until now is 267
19
+ 2022-02-20 12:02:06,876 INFO [decode.py:346] batch 20/?, cuts processed until now is 281
20
+ 2022-02-20 12:02:28,941 INFO [decode.py:346] batch 22/?, cuts processed until now is 325
21
+ 2022-02-20 12:02:51,934 INFO [decode.py:346] batch 24/?, cuts processed until now is 349
22
+ 2022-02-20 12:03:16,462 INFO [decode.py:346] batch 26/?, cuts processed until now is 366
23
+ 2022-02-20 12:03:38,858 INFO [decode.py:346] batch 28/?, cuts processed until now is 389
24
+ 2022-02-20 12:04:01,518 INFO [decode.py:346] batch 30/?, cuts processed until now is 410
25
+ 2022-02-20 12:04:23,552 INFO [decode.py:346] batch 32/?, cuts processed until now is 442
26
+ 2022-02-20 12:04:46,455 INFO [decode.py:346] batch 34/?, cuts processed until now is 474
27
+ 2022-02-20 12:05:09,016 INFO [decode.py:346] batch 36/?, cuts processed until now is 498
28
+ 2022-02-20 12:05:32,340 INFO [decode.py:346] batch 38/?, cuts processed until now is 514
29
+ 2022-02-20 12:05:56,709 INFO [decode.py:346] batch 40/?, cuts processed until now is 524
30
+ 2022-02-20 12:06:19,540 INFO [decode.py:346] batch 42/?, cuts processed until now is 564
31
+ 2022-02-20 12:06:41,687 INFO [decode.py:346] batch 44/?, cuts processed until now is 621
32
+ 2022-02-20 12:07:05,134 INFO [decode.py:346] batch 46/?, cuts processed until now is 633
33
+ 2022-02-20 12:07:27,945 INFO [decode.py:346] batch 48/?, cuts processed until now is 648
34
+ 2022-02-20 12:07:50,752 INFO [decode.py:346] batch 50/?, cuts processed until now is 662
35
+ 2022-02-20 12:08:13,070 INFO [decode.py:346] batch 52/?, cuts processed until now is 680
36
+ 2022-02-20 12:08:35,388 INFO [decode.py:346] batch 54/?, cuts processed until now is 708
37
+ 2022-02-20 12:08:57,030 INFO [decode.py:346] batch 56/?, cuts processed until now is 736
38
+ 2022-02-20 12:09:19,280 INFO [decode.py:346] batch 58/?, cuts processed until now is 752
39
+ 2022-02-20 12:09:41,384 INFO [decode.py:346] batch 60/?, cuts processed until now is 819
40
+ 2022-02-20 12:10:03,918 INFO [decode.py:346] batch 62/?, cuts processed until now is 843
41
+ 2022-02-20 12:10:28,563 INFO [decode.py:346] batch 64/?, cuts processed until now is 850
42
+ 2022-02-20 12:10:50,611 INFO [decode.py:346] batch 66/?, cuts processed until now is 880
43
+ 2022-02-20 12:11:13,026 INFO [decode.py:346] batch 68/?, cuts processed until now is 932
44
+ 2022-02-20 12:11:37,532 INFO [decode.py:346] batch 70/?, cuts processed until now is 957
45
+ 2022-02-20 12:11:59,446 INFO [decode.py:346] batch 72/?, cuts processed until now is 996
46
+ 2022-02-20 12:12:21,767 INFO [decode.py:346] batch 74/?, cuts processed until now is 1064
47
+ 2022-02-20 12:12:45,513 INFO [decode.py:346] batch 76/?, cuts processed until now is 1077
48
+ 2022-02-20 12:13:08,832 INFO [decode.py:346] batch 78/?, cuts processed until now is 1114
49
+ 2022-02-20 12:13:30,964 INFO [decode.py:346] batch 80/?, cuts processed until now is 1141
50
+ 2022-02-20 12:13:52,750 INFO [decode.py:346] batch 82/?, cuts processed until now is 1172
51
+ 2022-02-20 12:14:17,953 INFO [decode.py:346] batch 84/?, cuts processed until now is 1180
52
+ 2022-02-20 12:14:41,071 INFO [decode.py:346] batch 86/?, cuts processed until now is 1199
53
+ 2022-02-20 12:15:04,109 INFO [decode.py:346] batch 88/?, cuts processed until now is 1228
54
+ 2022-02-20 12:15:28,187 INFO [decode.py:346] batch 90/?, cuts processed until now is 1245
55
+ 2022-02-20 12:15:51,290 INFO [decode.py:346] batch 92/?, cuts processed until now is 1267
56
+ 2022-02-20 12:16:14,383 INFO [decode.py:346] batch 94/?, cuts processed until now is 1333
57
+ 2022-02-20 12:16:37,233 INFO [decode.py:346] batch 96/?, cuts processed until now is 1363
58
+ 2022-02-20 12:16:49,981 INFO [decode.py:346] batch 98/?, cuts processed until now is 1378
59
+ 2022-02-20 12:17:12,524 INFO [decode.py:346] batch 100/?, cuts processed until now is 1406
60
+ 2022-02-20 12:17:35,452 INFO [decode.py:346] batch 102/?, cuts processed until now is 1420
61
+ 2022-02-20 12:17:57,850 INFO [decode.py:346] batch 104/?, cuts processed until now is 1445
62
+ 2022-02-20 12:18:20,791 INFO [decode.py:346] batch 106/?, cuts processed until now is 1468
63
+ 2022-02-20 12:18:44,009 INFO [decode.py:346] batch 108/?, cuts processed until now is 1494
64
+ 2022-02-20 12:19:06,192 INFO [decode.py:346] batch 110/?, cuts processed until now is 1547
65
+ 2022-02-20 12:19:29,548 INFO [decode.py:346] batch 112/?, cuts processed until now is 1582
66
+ 2022-02-20 12:19:51,940 INFO [decode.py:346] batch 114/?, cuts processed until now is 1615
67
+ 2022-02-20 12:20:15,496 INFO [decode.py:346] batch 116/?, cuts processed until now is 1636
68
+ 2022-02-20 12:20:37,647 INFO [decode.py:346] batch 118/?, cuts processed until now is 1655
69
+ 2022-02-20 12:21:00,366 INFO [decode.py:346] batch 120/?, cuts processed until now is 1696
70
+ 2022-02-20 12:21:22,837 INFO [decode.py:346] batch 122/?, cuts processed until now is 1722
71
+ 2022-02-20 12:21:46,075 INFO [decode.py:346] batch 124/?, cuts processed until now is 1734
72
+ 2022-02-20 12:22:09,006 INFO [decode.py:346] batch 126/?, cuts processed until now is 1755
73
+ 2022-02-20 12:22:31,043 INFO [decode.py:346] batch 128/?, cuts processed until now is 1781
74
+ 2022-02-20 12:22:52,852 INFO [decode.py:346] batch 130/?, cuts processed until now is 1844
75
+ 2022-02-20 12:23:14,742 INFO [decode.py:346] batch 132/?, cuts processed until now is 1886
76
+ 2022-02-20 12:23:36,721 INFO [decode.py:346] batch 134/?, cuts processed until now is 1903
77
+ 2022-02-20 12:23:58,545 INFO [decode.py:346] batch 136/?, cuts processed until now is 1932
78
+ 2022-02-20 12:24:22,161 INFO [decode.py:346] batch 138/?, cuts processed until now is 1951
79
+ 2022-02-20 12:24:43,853 INFO [decode.py:346] batch 140/?, cuts processed until now is 1959
80
+ 2022-02-20 12:25:05,740 INFO [decode.py:346] batch 142/?, cuts processed until now is 2007
81
+ 2022-02-20 12:25:27,909 INFO [decode.py:346] batch 144/?, cuts processed until now is 2035
82
+ 2022-02-20 12:25:49,362 INFO [decode.py:346] batch 146/?, cuts processed until now is 2083
83
+ 2022-02-20 12:26:08,990 INFO [decode.py:346] batch 148/?, cuts processed until now is 2130
84
+ 2022-02-20 12:26:32,813 INFO [decode.py:346] batch 150/?, cuts processed until now is 2139
85
+ 2022-02-20 12:26:55,356 INFO [decode.py:346] batch 152/?, cuts processed until now is 2171
86
+ 2022-02-20 12:27:19,489 INFO [decode.py:346] batch 154/?, cuts processed until now is 2192
87
+ 2022-02-20 12:27:44,236 INFO [decode.py:346] batch 156/?, cuts processed until now is 2208
88
+ 2022-02-20 12:28:07,347 INFO [decode.py:346] batch 158/?, cuts processed until now is 2230
89
+ 2022-02-20 12:28:31,364 INFO [decode.py:346] batch 160/?, cuts processed until now is 2258
90
+ 2022-02-20 12:28:56,591 INFO [decode.py:346] batch 162/?, cuts processed until now is 2274
91
+ 2022-02-20 12:29:19,729 INFO [decode.py:346] batch 164/?, cuts processed until now is 2296
92
+ 2022-02-20 12:29:42,203 INFO [decode.py:346] batch 166/?, cuts processed until now is 2327
93
+ 2022-02-20 12:30:04,439 INFO [decode.py:346] batch 168/?, cuts processed until now is 2351
94
+ 2022-02-20 12:30:28,200 INFO [decode.py:346] batch 170/?, cuts processed until now is 2364
95
+ 2022-02-20 12:30:52,497 INFO [decode.py:346] batch 172/?, cuts processed until now is 2374
96
+ 2022-02-20 12:31:15,224 INFO [decode.py:346] batch 174/?, cuts processed until now is 2399
97
+ 2022-02-20 12:31:38,136 INFO [decode.py:346] batch 176/?, cuts processed until now is 2418
98
+ 2022-02-20 12:31:52,073 INFO [decode.py:346] batch 178/?, cuts processed until now is 2427
99
+ 2022-02-20 12:32:14,714 INFO [decode.py:346] batch 180/?, cuts processed until now is 2440
100
+ 2022-02-20 12:32:37,402 INFO [decode.py:346] batch 182/?, cuts processed until now is 2456
101
+ 2022-02-20 12:32:59,728 INFO [decode.py:346] batch 184/?, cuts processed until now is 2466
102
+ 2022-02-20 12:33:21,645 INFO [decode.py:346] batch 186/?, cuts processed until now is 2479
103
+ 2022-02-20 12:33:35,394 INFO [decode.py:346] batch 188/?, cuts processed until now is 2492
104
+ 2022-02-20 12:33:50,981 INFO [decode.py:346] batch 190/?, cuts processed until now is 2502
105
+ 2022-02-20 12:34:15,001 INFO [decode.py:346] batch 192/?, cuts processed until now is 2520
106
+ 2022-02-20 12:34:28,291 INFO [decode.py:346] batch 194/?, cuts processed until now is 2527
107
+ 2022-02-20 12:34:53,777 INFO [decode.py:346] batch 196/?, cuts processed until now is 2539
108
+ 2022-02-20 12:35:18,599 INFO [decode.py:346] batch 198/?, cuts processed until now is 2551
109
+ 2022-02-20 12:35:43,025 INFO [decode.py:346] batch 200/?, cuts processed until now is 2563
110
+ 2022-02-20 12:36:06,613 INFO [decode.py:346] batch 202/?, cuts processed until now is 2575
111
+ 2022-02-20 12:36:30,140 INFO [decode.py:346] batch 204/?, cuts processed until now is 2587
112
+ 2022-02-20 12:36:53,833 INFO [decode.py:346] batch 206/?, cuts processed until now is 2599
113
+ 2022-02-20 12:37:17,581 INFO [decode.py:346] batch 208/?, cuts processed until now is 2611
114
+ 2022-02-20 12:37:34,891 INFO [decode.py:346] batch 210/?, cuts processed until now is 2620
115
+ 2022-02-20 12:37:34,995 INFO [decode.py:363] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/modified_beam_search/recogs-test-clean-beam_4-epoch-57-avg-17-beam-4.txt
116
+ 2022-02-20 12:37:35,076 INFO [utils.py:404] [test-clean-beam_4] %WER 6.31% [3315 / 52576, 365 ins, 308 del, 2642 sub ]
117
+ 2022-02-20 12:37:35,305 INFO [decode.py:376] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/modified_beam_search/errs-test-clean-beam_4-epoch-57-avg-17-beam-4.txt
118
+ 2022-02-20 12:37:35,306 INFO [decode.py:393]
119
+ For test-clean, WER of different settings are:
120
+ beam_4 6.31 best for test-clean
121
+
122
+ 2022-02-20 12:37:47,631 INFO [decode.py:346] batch 0/?, cuts processed until now is 23
123
+ 2022-02-20 12:38:11,365 INFO [decode.py:346] batch 2/?, cuts processed until now is 71
124
+ 2022-02-20 12:38:35,898 INFO [decode.py:346] batch 4/?, cuts processed until now is 92
125
+ 2022-02-20 12:38:57,899 INFO [decode.py:346] batch 6/?, cuts processed until now is 134
126
+ 2022-02-20 12:39:22,523 INFO [decode.py:346] batch 8/?, cuts processed until now is 150
127
+ 2022-02-20 12:39:44,879 INFO [decode.py:346] batch 10/?, cuts processed until now is 190
128
+ 2022-02-20 12:40:06,441 INFO [decode.py:346] batch 12/?, cuts processed until now is 225
129
+ 2022-02-20 12:40:27,521 INFO [decode.py:346] batch 14/?, cuts processed until now is 240
130
+ 2022-02-20 12:40:49,397 INFO [decode.py:346] batch 16/?, cuts processed until now is 266
131
+ 2022-02-20 12:41:12,761 INFO [decode.py:346] batch 18/?, cuts processed until now is 296
132
+ 2022-02-20 12:41:37,753 INFO [decode.py:346] batch 20/?, cuts processed until now is 312
133
+ 2022-02-20 12:42:01,655 INFO [decode.py:346] batch 22/?, cuts processed until now is 360
134
+ 2022-02-20 12:42:25,393 INFO [decode.py:346] batch 24/?, cuts processed until now is 387
135
+ 2022-02-20 12:42:52,350 INFO [decode.py:346] batch 26/?, cuts processed until now is 406
136
+ 2022-02-20 12:43:16,218 INFO [decode.py:346] batch 28/?, cuts processed until now is 433
137
+ 2022-02-20 12:43:39,935 INFO [decode.py:346] batch 30/?, cuts processed until now is 457
138
+ 2022-02-20 12:44:03,561 INFO [decode.py:346] batch 32/?, cuts processed until now is 494
139
+ 2022-02-20 12:44:26,043 INFO [decode.py:346] batch 34/?, cuts processed until now is 540
140
+ 2022-02-20 12:44:48,809 INFO [decode.py:346] batch 36/?, cuts processed until now is 568
141
+ 2022-02-20 12:45:11,735 INFO [decode.py:346] batch 38/?, cuts processed until now is 587
142
+ 2022-02-20 12:45:35,826 INFO [decode.py:346] batch 40/?, cuts processed until now is 598
143
+ 2022-02-20 12:45:58,761 INFO [decode.py:346] batch 42/?, cuts processed until now is 640
144
+ 2022-02-20 12:46:20,759 INFO [decode.py:346] batch 44/?, cuts processed until now is 700
145
+ 2022-02-20 12:46:46,322 INFO [decode.py:346] batch 46/?, cuts processed until now is 715
146
+ 2022-02-20 12:47:10,371 INFO [decode.py:346] batch 48/?, cuts processed until now is 733
147
+ 2022-02-20 12:47:32,768 INFO [decode.py:346] batch 50/?, cuts processed until now is 749
148
+ 2022-02-20 12:47:57,053 INFO [decode.py:346] batch 52/?, cuts processed until now is 769
149
+ 2022-02-20 12:48:21,208 INFO [decode.py:346] batch 54/?, cuts processed until now is 801
150
+ 2022-02-20 12:48:44,960 INFO [decode.py:346] batch 56/?, cuts processed until now is 833
151
+ 2022-02-20 12:49:08,278 INFO [decode.py:346] batch 58/?, cuts processed until now is 853
152
+ 2022-02-20 12:49:30,111 INFO [decode.py:346] batch 60/?, cuts processed until now is 922
153
+ 2022-02-20 12:49:52,765 INFO [decode.py:346] batch 62/?, cuts processed until now is 950
154
+ 2022-02-20 12:50:15,818 INFO [decode.py:346] batch 64/?, cuts processed until now is 958
155
+ 2022-02-20 12:50:39,070 INFO [decode.py:346] batch 66/?, cuts processed until now is 992
156
+ 2022-02-20 12:51:02,031 INFO [decode.py:346] batch 68/?, cuts processed until now is 1046
157
+ 2022-02-20 12:51:24,412 INFO [decode.py:346] batch 70/?, cuts processed until now is 1074
158
+ 2022-02-20 12:51:46,826 INFO [decode.py:346] batch 72/?, cuts processed until now is 1119
159
+ 2022-02-20 12:52:08,835 INFO [decode.py:346] batch 74/?, cuts processed until now is 1188
160
+ 2022-02-20 12:52:30,909 INFO [decode.py:346] batch 76/?, cuts processed until now is 1203
161
+ 2022-02-20 12:52:55,192 INFO [decode.py:346] batch 78/?, cuts processed until now is 1243
162
+ 2022-02-20 12:53:17,525 INFO [decode.py:346] batch 80/?, cuts processed until now is 1274
163
+ 2022-02-20 12:53:39,837 INFO [decode.py:346] batch 82/?, cuts processed until now is 1310
164
+ 2022-02-20 12:54:05,422 INFO [decode.py:346] batch 84/?, cuts processed until now is 1320
165
+ 2022-02-20 12:54:28,259 INFO [decode.py:346] batch 86/?, cuts processed until now is 1343
166
+ 2022-02-20 12:54:50,897 INFO [decode.py:346] batch 88/?, cuts processed until now is 1377
167
+ 2022-02-20 12:55:13,516 INFO [decode.py:346] batch 90/?, cuts processed until now is 1422
168
+ 2022-02-20 12:55:36,689 INFO [decode.py:346] batch 92/?, cuts processed until now is 1448
169
+ 2022-02-20 12:56:00,164 INFO [decode.py:346] batch 94/?, cuts processed until now is 1514
170
+ 2022-02-20 12:56:24,228 INFO [decode.py:346] batch 96/?, cuts processed until now is 1547
171
+ 2022-02-20 12:56:43,524 INFO [decode.py:346] batch 98/?, cuts processed until now is 1594
172
+ 2022-02-20 12:57:06,552 INFO [decode.py:346] batch 100/?, cuts processed until now is 1614
173
+ 2022-02-20 12:57:31,323 INFO [decode.py:346] batch 102/?, cuts processed until now is 1631
174
+ 2022-02-20 12:57:55,007 INFO [decode.py:346] batch 104/?, cuts processed until now is 1660
175
+ 2022-02-20 12:58:18,436 INFO [decode.py:346] batch 106/?, cuts processed until now is 1686
176
+ 2022-02-20 12:58:41,830 INFO [decode.py:346] batch 108/?, cuts processed until now is 1716
177
+ 2022-02-20 12:59:05,013 INFO [decode.py:346] batch 110/?, cuts processed until now is 1774
178
+ 2022-02-20 12:59:28,064 INFO [decode.py:346] batch 112/?, cuts processed until now is 1812
179
+ 2022-02-20 12:59:51,705 INFO [decode.py:346] batch 114/?, cuts processed until now is 1849
180
+ 2022-02-20 13:00:14,228 INFO [decode.py:346] batch 116/?, cuts processed until now is 1872
181
+ 2022-02-20 13:00:37,211 INFO [decode.py:346] batch 118/?, cuts processed until now is 1894
182
+ 2022-02-20 13:00:59,514 INFO [decode.py:346] batch 120/?, cuts processed until now is 1938
183
+ 2022-02-20 13:01:22,345 INFO [decode.py:346] batch 122/?, cuts processed until now is 1968
184
+ 2022-02-20 13:01:45,374 INFO [decode.py:346] batch 124/?, cuts processed until now is 1982
185
+ 2022-02-20 13:02:08,660 INFO [decode.py:346] batch 126/?, cuts processed until now is 2007
186
+ 2022-02-20 13:02:30,940 INFO [decode.py:346] batch 128/?, cuts processed until now is 2037
187
+ 2022-02-20 13:02:52,930 INFO [decode.py:346] batch 130/?, cuts processed until now is 2105
188
+ 2022-02-20 13:03:14,875 INFO [decode.py:346] batch 132/?, cuts processed until now is 2152
189
+ 2022-02-20 13:03:37,414 INFO [decode.py:346] batch 134/?, cuts processed until now is 2172
190
+ 2022-02-20 13:03:59,889 INFO [decode.py:346] batch 136/?, cuts processed until now is 2206
191
+ 2022-02-20 13:04:23,198 INFO [decode.py:346] batch 138/?, cuts processed until now is 2228
192
+ 2022-02-20 13:04:45,786 INFO [decode.py:346] batch 140/?, cuts processed until now is 2238
193
+ 2022-02-20 13:05:07,929 INFO [decode.py:346] batch 142/?, cuts processed until now is 2290
194
+ 2022-02-20 13:05:30,295 INFO [decode.py:346] batch 144/?, cuts processed until now is 2321
195
+ 2022-02-20 13:05:46,062 INFO [decode.py:346] batch 146/?, cuts processed until now is 2353
196
+ 2022-02-20 13:06:08,028 INFO [decode.py:346] batch 148/?, cuts processed until now is 2410
197
+ 2022-02-20 13:06:29,925 INFO [decode.py:346] batch 150/?, cuts processed until now is 2420
198
+ 2022-02-20 13:06:52,188 INFO [decode.py:346] batch 152/?, cuts processed until now is 2456
199
+ 2022-02-20 13:07:14,028 INFO [decode.py:346] batch 154/?, cuts processed until now is 2479
200
+ 2022-02-20 13:07:36,102 INFO [decode.py:346] batch 156/?, cuts processed until now is 2496
201
+ 2022-02-20 13:07:50,292 INFO [decode.py:346] batch 158/?, cuts processed until now is 2513
202
+ 2022-02-20 13:08:14,376 INFO [decode.py:346] batch 160/?, cuts processed until now is 2531
203
+ 2022-02-20 13:08:36,443 INFO [decode.py:346] batch 162/?, cuts processed until now is 2555
204
+ 2022-02-20 13:08:59,094 INFO [decode.py:346] batch 164/?, cuts processed until now is 2587
205
+ 2022-02-20 13:09:20,773 INFO [decode.py:346] batch 166/?, cuts processed until now is 2621
206
+ 2022-02-20 13:09:43,489 INFO [decode.py:346] batch 168/?, cuts processed until now is 2650
207
+ 2022-02-20 13:09:58,731 INFO [decode.py:346] batch 170/?, cuts processed until now is 2661
208
+ 2022-02-20 13:10:23,385 INFO [decode.py:346] batch 172/?, cuts processed until now is 2673
209
+ 2022-02-20 13:10:46,807 INFO [decode.py:346] batch 174/?, cuts processed until now is 2698
210
+ 2022-02-20 13:11:09,942 INFO [decode.py:346] batch 176/?, cuts processed until now is 2721
211
+ 2022-02-20 13:11:33,462 INFO [decode.py:346] batch 178/?, cuts processed until now is 2737
212
+ 2022-02-20 13:11:49,551 INFO [decode.py:346] batch 180/?, cuts processed until now is 2758
213
+ 2022-02-20 13:12:13,168 INFO [decode.py:346] batch 182/?, cuts processed until now is 2770
214
+ 2022-02-20 13:12:35,966 INFO [decode.py:346] batch 184/?, cuts processed until now is 2786
215
+ 2022-02-20 13:12:57,323 INFO [decode.py:346] batch 186/?, cuts processed until now is 2805
216
+ 2022-02-20 13:13:19,945 INFO [decode.py:346] batch 188/?, cuts processed until now is 2817
217
+ 2022-02-20 13:13:32,237 INFO [decode.py:346] batch 190/?, cuts processed until now is 2825
218
+ 2022-02-20 13:13:51,907 INFO [decode.py:346] batch 192/?, cuts processed until now is 2843
219
+ 2022-02-20 13:14:16,910 INFO [decode.py:346] batch 194/?, cuts processed until now is 2857
220
+ 2022-02-20 13:14:41,212 INFO [decode.py:346] batch 196/?, cuts processed until now is 2871
221
+ 2022-02-20 13:15:05,033 INFO [decode.py:346] batch 198/?, cuts processed until now is 2885
222
+ 2022-02-20 13:15:28,269 INFO [decode.py:346] batch 200/?, cuts processed until now is 2899
223
+ 2022-02-20 13:15:51,046 INFO [decode.py:346] batch 202/?, cuts processed until now is 2913
224
+ 2022-02-20 13:16:13,493 INFO [decode.py:346] batch 204/?, cuts processed until now is 2927
225
+ 2022-02-20 13:16:32,234 INFO [decode.py:346] batch 206/?, cuts processed until now is 2939
226
+ 2022-02-20 13:16:32,360 INFO [decode.py:363] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/modified_beam_search/recogs-test-other-beam_4-epoch-57-avg-17-beam-4.txt
227
+ 2022-02-20 13:16:32,445 INFO [utils.py:404] [test-other-beam_4] %WER 16.30% [8531 / 52343, 860 ins, 936 del, 6735 sub ]
228
+ 2022-02-20 13:16:32,689 INFO [decode.py:376] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/modified_beam_search/errs-test-other-beam_4-epoch-57-avg-17-beam-4.txt
229
+ 2022-02-20 13:16:32,690 INFO [decode.py:393]
230
+ For test-other, WER of different settings are:
231
+ beam_4 16.3 best for test-other
232
+
233
+ 2022-02-20 13:16:32,690 INFO [decode.py:488] Done!
log/log-decode-epoch-57-avg-17-context-2-max-sym-per-frame-1-2022-02-20-11-27-46 ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-02-20 11:27:46,890 INFO [decode.py:421] Decoding started
2
+ 2022-02-20 11:27:46,890 INFO [decode.py:427] Device: cuda:0
3
+ 2022-02-20 11:27:46,892 INFO [decode.py:436] {'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': 'f4fefe4882bc0ae59af951da3f47335d5495ef71', 'k2-git-date': 'Thu Feb 10 15:16:02 2022', 'lhotse-version': '1.0.0.dev+missing.version.file', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'multiple-datasets', 'icefall-git-sha1': '61b0019-dirty', 'icefall-git-date': 'Thu Feb 17 18:34:48 2022', 'icefall-path': '/ceph-fj/fangjun/open-source-2/icefall-multi-datasets', 'k2-path': '/ceph-fj/fangjun/open-source-2/k2-multi-datasets/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-fj/fangjun/open-source-2/lhotse-multi-datasets/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-5-0816110343-9647676d5-sqn62', 'IP address': '10.177.63.199'}, 'epoch': 57, 'avg': 17, 'exp_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2'), 'bpe_model': './data/lang_bpe_500/bpe.model', 'decoding_method': 'greedy_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 1, 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'manifest_dir': PosixPath('data/fbank'), 'on_the_fly_feats': False, 'res_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2/greedy_search'), 'suffix': 'epoch-57-avg-17-context-2-max-sym-per-frame-1', 'blank_id': 0, 'vocab_size': 500}
4
+ 2022-02-20 11:27:46,892 INFO [decode.py:438] About to create model
5
+ 2022-02-20 11:27:47,391 INFO [decode.py:449] averaging ['transducer_stateless_multi_datasets/exp-100-2/epoch-41.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-42.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-43.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-44.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-45.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-46.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-47.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-48.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-49.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-50.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-51.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-52.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-53.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-54.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-55.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-56.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-57.pt']
6
+ 2022-02-20 11:28:09,571 INFO [decode.py:460] Number of model parameters: 84521448
7
+ 2022-02-20 11:28:09,572 INFO [librispeech.py:58] About to get test-clean cuts from data/fbank/cuts_test-clean.json.gz
8
+ 2022-02-20 11:28:09,793 INFO [librispeech.py:63] About to get test-other cuts from data/fbank/cuts_test-other.json.gz
9
+ 2022-02-20 11:28:11,190 INFO [decode.py:346] batch 0/?, cuts processed until now is 20
10
+ 2022-02-20 11:29:35,058 INFO [decode.py:346] batch 100/?, cuts processed until now is 1406
11
+ 2022-02-20 11:30:55,213 INFO [decode.py:346] batch 200/?, cuts processed until now is 2563
12
+ 2022-02-20 11:31:02,810 INFO [decode.py:363] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/greedy_search/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt
13
+ 2022-02-20 11:31:02,884 INFO [utils.py:404] [test-clean-greedy_search] %WER 6.34% [3333 / 52576, 351 ins, 330 del, 2652 sub ]
14
+ 2022-02-20 11:31:03,106 INFO [decode.py:376] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/greedy_search/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt
15
+ 2022-02-20 11:31:03,106 INFO [decode.py:393]
16
+ For test-clean, WER of different settings are:
17
+ greedy_search 6.34 best for test-clean
18
+
19
+ 2022-02-20 11:31:04,247 INFO [decode.py:346] batch 0/?, cuts processed until now is 23
20
+ 2022-02-20 11:32:29,360 INFO [decode.py:346] batch 100/?, cuts processed until now is 1614
21
+ 2022-02-20 11:33:36,657 INFO [decode.py:346] batch 200/?, cuts processed until now is 2899
22
+ 2022-02-20 11:33:41,222 INFO [decode.py:363] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/greedy_search/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt
23
+ 2022-02-20 11:33:41,301 INFO [utils.py:404] [test-other-greedy_search] %WER 16.70% [8740 / 52343, 838 ins, 1067 del, 6835 sub ]
24
+ 2022-02-20 11:33:41,540 INFO [decode.py:376] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/greedy_search/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt
25
+ 2022-02-20 11:33:41,540 INFO [decode.py:393]
26
+ For test-other, WER of different settings are:
27
+ greedy_search 16.7 best for test-other
28
+
29
+ 2022-02-20 11:33:41,540 INFO [decode.py:488] Done!
log/log-decode-epoch-57-avg-17-context-2-max-sym-per-frame-2-2022-02-21-11-31-01 ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-02-21 11:31:01,353 INFO [decode.py:421] Decoding started
2
+ 2022-02-21 11:31:01,353 INFO [decode.py:427] Device: cuda:0
3
+ 2022-02-21 11:31:01,360 INFO [decode.py:436] {'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': 'f4fefe4882bc0ae59af951da3f47335d5495ef71', 'k2-git-date': 'Thu Feb 10 15:16:02 2022', 'lhotse-version': '1.0.0.dev+missing.version.file', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'multiple-datasets', 'icefall-git-sha1': '61b0019-dirty', 'icefall-git-date': 'Thu Feb 17 18:34:48 2022', 'icefall-path': '/ceph-fj/fangjun/open-source-2/icefall-multi-datasets', 'k2-path': '/ceph-fj/fangjun/open-source-2/k2-multi-datasets/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-fj/fangjun/open-source-2/lhotse-multi-datasets/lhotse/__init__.py', 'hostname': 'de-74279-k2-dev-1-0203142625-9776c46db-pk7w6', 'IP address': '10.177.22.139'}, 'epoch': 57, 'avg': 17, 'exp_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2'), 'bpe_model': './data/lang_bpe_500/bpe.model', 'decoding_method': 'greedy_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 2, 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'manifest_dir': PosixPath('data/fbank'), 'on_the_fly_feats': False, 'res_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2/greedy_search'), 'suffix': 'epoch-57-avg-17-context-2-max-sym-per-frame-2', 'blank_id': 0, 'vocab_size': 500}
4
+ 2022-02-21 11:31:01,360 INFO [decode.py:438] About to create model
5
+ 2022-02-21 11:31:01,990 INFO [decode.py:449] averaging ['transducer_stateless_multi_datasets/exp-100-2/epoch-41.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-42.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-43.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-44.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-45.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-46.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-47.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-48.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-49.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-50.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-51.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-52.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-53.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-54.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-55.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-56.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-57.pt']
6
+ 2022-02-21 11:32:44,770 INFO [decode.py:460] Number of model parameters: 84521448
7
+ 2022-02-21 11:32:44,770 INFO [librispeech.py:58] About to get test-clean cuts from data/fbank/cuts_test-clean.json.gz
8
+ 2022-02-21 11:32:44,994 INFO [librispeech.py:63] About to get test-other cuts from data/fbank/cuts_test-other.json.gz
9
+ 2022-02-21 11:32:47,120 INFO [decode.py:346] batch 0/?, cuts processed until now is 20
10
+ 2022-02-21 11:35:12,670 INFO [decode.py:346] batch 100/?, cuts processed until now is 1406
11
+ 2022-02-21 11:37:35,107 INFO [decode.py:346] batch 200/?, cuts processed until now is 2563
12
+ 2022-02-21 11:37:48,512 INFO [decode.py:363] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/greedy_search/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt
13
+ 2022-02-21 11:37:48,609 INFO [utils.py:404] [test-clean-greedy_search] %WER 6.34% [3331 / 52576, 352 ins, 328 del, 2651 sub ]
14
+ 2022-02-21 11:37:48,921 INFO [decode.py:376] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/greedy_search/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt
15
+ 2022-02-21 11:37:48,922 INFO [decode.py:393]
16
+ For test-clean, WER of different settings are:
17
+ greedy_search 6.34 best for test-clean
18
+
19
+ 2022-02-21 11:37:50,838 INFO [decode.py:346] batch 0/?, cuts processed until now is 23
20
+ 2022-02-21 11:40:17,151 INFO [decode.py:346] batch 100/?, cuts processed until now is 1614
21
+ 2022-02-21 11:42:30,546 INFO [decode.py:346] batch 200/?, cuts processed until now is 2899
22
+ 2022-02-21 11:42:38,702 INFO [decode.py:363] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/greedy_search/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt
23
+ 2022-02-21 11:42:38,808 INFO [utils.py:404] [test-other-greedy_search] %WER 16.70% [8740 / 52343, 841 ins, 1064 del, 6835 sub ]
24
+ 2022-02-21 11:42:39,142 INFO [decode.py:376] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/greedy_search/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt
25
+ 2022-02-21 11:42:39,144 INFO [decode.py:393]
26
+ For test-other, WER of different settings are:
27
+ greedy_search 16.7 best for test-other
28
+
29
+ 2022-02-21 11:42:39,144 INFO [decode.py:488] Done!
log/log-decode-epoch-57-avg-17-context-2-max-sym-per-frame-3-2022-02-21-12-13-11 ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-02-21 12:13:11,351 INFO [decode.py:416] Decoding started
2
+ 2022-02-21 12:13:11,351 INFO [decode.py:422] Device: cuda:0
3
+ 2022-02-21 12:13:11,355 INFO [decode.py:431] {'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': 'f4fefe4882bc0ae59af951da3f47335d5495ef71', 'k2-git-date': 'Thu Feb 10 15:16:02 2022', 'lhotse-version': '1.0.0.dev+missing.version.file', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'multiple-datasets', 'icefall-git-sha1': '61b0019-dirty', 'icefall-git-date': 'Thu Feb 17 18:34:48 2022', 'icefall-path': '/ceph-fj/fangjun/open-source-2/icefall-multi-datasets', 'k2-path': '/ceph-fj/fangjun/open-source-2/k2-multi-datasets/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-fj/fangjun/open-source-2/lhotse-multi-datasets/lhotse/__init__.py', 'hostname': 'de-74279-k2-dev-1-0203142625-9776c46db-pk7w6', 'IP address': '10.177.22.139'}, 'epoch': 57, 'avg': 17, 'exp_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2'), 'bpe_model': './data/lang_bpe_500/bpe.model', 'decoding_method': 'greedy_search', 'beam_size': 4, 'context_size': 2, 'max_sym_per_frame': 3, 'max_duration': 100, 'bucketing_sampler': True, 'num_buckets': 30, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'manifest_dir': PosixPath('data/fbank'), 'on_the_fly_feats': False, 'res_dir': PosixPath('transducer_stateless_multi_datasets/exp-100-2/greedy_search'), 'suffix': 'epoch-57-avg-17-context-2-max-sym-per-frame-3', 'blank_id': 0, 'vocab_size': 500}
4
+ 2022-02-21 12:13:11,355 INFO [decode.py:433] About to create model
5
+ 2022-02-21 12:13:11,967 INFO [decode.py:444] averaging ['transducer_stateless_multi_datasets/exp-100-2/epoch-41.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-42.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-43.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-44.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-45.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-46.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-47.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-48.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-49.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-50.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-51.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-52.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-53.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-54.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-55.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-56.pt', 'transducer_stateless_multi_datasets/exp-100-2/epoch-57.pt']
6
+ 2022-02-21 12:14:44,692 INFO [decode.py:455] Number of model parameters: 84007924
7
+ 2022-02-21 12:14:44,692 INFO [librispeech.py:58] About to get test-clean cuts from data/fbank/cuts_test-clean.json.gz
8
+ 2022-02-21 12:14:44,916 INFO [librispeech.py:63] About to get test-other cuts from data/fbank/cuts_test-other.json.gz
9
+ 2022-02-21 12:14:46,998 INFO [decode.py:341] batch 0/?, cuts processed until now is 20
10
+ 2022-02-21 12:17:12,777 INFO [decode.py:341] batch 100/?, cuts processed until now is 1406
11
+ 2022-02-21 12:19:27,830 INFO [decode.py:341] batch 200/?, cuts processed until now is 2563
12
+ 2022-02-21 12:19:41,254 INFO [decode.py:358] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/greedy_search/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt
13
+ 2022-02-21 12:19:41,371 INFO [utils.py:404] [test-clean-greedy_search] %WER 6.34% [3331 / 52576, 352 ins, 328 del, 2651 sub ]
14
+ 2022-02-21 12:19:41,679 INFO [decode.py:371] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/greedy_search/errs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt
15
+ 2022-02-21 12:19:41,680 INFO [decode.py:388]
16
+ For test-clean, WER of different settings are:
17
+ greedy_search 6.34 best for test-clean
18
+
19
+ 2022-02-21 12:19:43,450 INFO [decode.py:341] batch 0/?, cuts processed until now is 23
20
+ 2022-02-21 12:22:09,181 INFO [decode.py:341] batch 100/?, cuts processed until now is 1614
21
+ 2022-02-21 12:24:30,242 INFO [decode.py:341] batch 200/?, cuts processed until now is 2899
22
+ 2022-02-21 12:24:38,394 INFO [decode.py:358] The transcripts are stored in transducer_stateless_multi_datasets/exp-100-2/greedy_search/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt
23
+ 2022-02-21 12:24:38,500 INFO [utils.py:404] [test-other-greedy_search] %WER 16.70% [8739 / 52343, 841 ins, 1064 del, 6834 sub ]
24
+ 2022-02-21 12:24:38,843 INFO [decode.py:371] Wrote detailed error stats to transducer_stateless_multi_datasets/exp-100-2/greedy_search/errs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt
25
+ 2022-02-21 12:24:38,844 INFO [decode.py:388]
26
+ For test-other, WER of different settings are:
27
+ greedy_search 16.7 best for test-other
28
+
29
+ 2022-02-21 12:24:38,844 INFO [decode.py:483] Done!
log/recogs-test-clean-beam_4-epoch-57-avg-17-beam-4.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/recogs-test-clean-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/recogs-test-other-beam_4-epoch-57-avg-17-beam-4.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-1.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-2.txt ADDED
The diff for this file is too large to render. See raw diff
 
log/recogs-test-other-greedy_search-epoch-57-avg-17-context-2-max-sym-per-frame-3.txt ADDED
The diff for this file is too large to render. See raw diff
 
test_wavs/1089-134686-0001.wav ADDED
Binary file (212 kB). View file
 
test_wavs/1221-135766-0001.wav ADDED
Binary file (535 kB). View file
 
test_wavs/1221-135766-0002.wav ADDED
Binary file (154 kB). View file
 
test_wavs/trans.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ 1089-134686-0001 AFTER EARLY NIGHTFALL THE YELLOW LAMPS WOULD LIGHT UP HERE AND THERE THE SQUALID QUARTER OF THE BROTHELS
2
+ 1221-135766-0001 GOD AS A DIRECT CONSEQUENCE OF THE SIN WHICH MAN THUS PUNISHED HAD GIVEN HER A LOVELY CHILD WHOSE PLACE WAS ON THAT SAME DISHONOURED BOSOM TO CONNECT HER PARENT FOR EVER WITH THE RACE AND DESCENT OF MORTALS AND TO BE FINALLY A BLESSED SOUL IN HEAVEN
3
+ 1221-135766-0002 YET THESE THOUGHTS AFFECTED HESTER PRYNNE LESS WITH HOPE THAN APPREHENSION