Hugging Face's logo

desh2608
/
icefall-asr-alimeeting-pruned-transducer-stateless7

K2
TensorBoard
Chinese
icefall
Model card Files Metrics Community
2
icefall-asr-alimeeting-pruned-transducer-stateless7
File size: 15,114 Bytes 
26cafdd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
 
2022-12-08 23:52:09,512 INFO [decode.py:551] Decoding started
2022-12-08 23:52:09,513 INFO [decode.py:557] Device: cuda:0
2022-12-08 23:52:09,579 INFO [lexicon.py:168] Loading pre-compiled data/lang_char/Linv.pt
2022-12-08 23:52:09,589 INFO [decode.py:563] {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 100, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.23', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'b2ce63f3940018e7b433c43fd802fc50ab006a76', 'k2-git-date': 'Wed Nov 23 08:43:43 2022', 'lhotse-version': '1.9.0.dev+git.97bf4b0.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'ali_meeting', 'icefall-git-sha1': 'f13cf61-dirty', 'icefall-git-date': 'Tue Dec 6 03:34:27 2022', 'icefall-path': '/exp/draj/mini_scale_2022/icefall', 'k2-path': '/exp/draj/mini_scale_2022/k2/k2/python/k2/__init__.py', 'lhotse-path': '/exp/draj/mini_scale_2022/lhotse/lhotse/__init__.py', 'hostname': 'r2n06', 'IP address': '10.1.2.6'}, 'epoch': 15, 'iter': 0, 'avg': 8, 'use_averaged_model': True, 'exp_dir': PosixPath('pruned_transducer_stateless7/exp/v1'), 'lang_dir': 'data/lang_char', 'decoding_method': 'greedy_search', 'beam_size': 4, 'beam': 4, 'ngram_lm_scale': 0.01, 'max_contexts': 4, 'max_states': 8, 'context_size': 2, 'max_sym_per_frame': 1, 'num_paths': 200, 'nbest_scale': 0.5, 'num_encoder_layers': '2,4,3,2,4', 'feedforward_dims': '1024,1024,2048,2048,1024', 'nhead': '8,8,8,8,8', 'encoder_dims': '384,384,384,384,384', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '256,256,256,256,256', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'manifest_dir': PosixPath('data/manifests'), 'enable_musan': True, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'max_duration': 500, 'max_cuts': None, 'num_buckets': 50, 'on_the_fly_feats': False, 'shuffle': True, 'num_workers': 8, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'res_dir': PosixPath('pruned_transducer_stateless7/exp/v1/greedy_search'), 'suffix': 'epoch-15-avg-8-context-2-max-sym-per-frame-1', 'blank_id': 0, 'vocab_size': 3290}
2022-12-08 23:52:09,589 INFO [decode.py:565] About to create model
2022-12-08 23:52:10,047 INFO [zipformer.py:179] At encoder stack 4, which has downsampling_factor=2, we will combine the outputs of layers 1 and 3, with downsampling_factors=2 and 8.
2022-12-08 23:52:10,093 INFO [decode.py:632] Calculating the averaged model over epoch range from 7 (excluded) to 15
2022-12-08 23:52:26,211 INFO [decode.py:655] Number of model parameters: 75734561
2022-12-08 23:52:26,212 INFO [asr_datamodule.py:381] About to get AliMeeting IHM eval cuts
2022-12-08 23:52:26,214 INFO [asr_datamodule.py:402] About to get AliMeeting IHM test cuts
2022-12-08 23:52:26,216 INFO [asr_datamodule.py:387] About to get AliMeeting SDM eval cuts
2022-12-08 23:52:26,217 INFO [asr_datamodule.py:408] About to get AliMeeting SDM test cuts
2022-12-08 23:52:26,219 INFO [asr_datamodule.py:396] About to get AliMeeting GSS-enhanced eval cuts
2022-12-08 23:52:26,221 INFO [asr_datamodule.py:417] About to get AliMeeting GSS-enhanced test cuts
2022-12-08 23:52:27,975 INFO [decode.py:687] Decoding eval_ihm
2022-12-08 23:52:29,438 INFO [decode.py:463] batch 0/?, cuts processed until now is 58
2022-12-08 23:52:52,862 INFO [decode.py:479] The transcripts are stored in pruned_transducer_stateless7/exp/v1/greedy_search/recogs-eval_ihm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:52:52,958 INFO [utils.py:536] [eval_ihm-greedy_search] %WER 10.13% [8216 / 81111, 831 ins, 2185 del, 5200 sub ]
2022-12-08 23:52:53,196 INFO [decode.py:492] Wrote detailed error stats to pruned_transducer_stateless7/exp/v1/greedy_search/errs-eval_ihm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:52:53,197 INFO [decode.py:508] 
For eval_ihm, WER of different settings are:
greedy_search	10.13	best for eval_ihm

2022-12-08 23:52:53,197 INFO [decode.py:687] Decoding test_ihm
2022-12-08 23:52:54,874 INFO [decode.py:463] batch 0/?, cuts processed until now is 49
2022-12-08 23:53:30,263 INFO [zipformer.py:1414] attn_weights_entropy = tensor([4.5696, 4.6696, 4.8707, 4.0575, 4.6745, 4.9781, 2.0707, 4.4098],
       device='cuda:0'), covar=tensor([0.0117, 0.0196, 0.0221, 0.0396, 0.0183, 0.0096, 0.3094, 0.0230],
       device='cuda:0'), in_proj_covar=tensor([0.0143, 0.0153, 0.0125, 0.0123, 0.0184, 0.0118, 0.0149, 0.0172],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003],
       device='cuda:0')
2022-12-08 23:53:49,853 INFO [zipformer.py:1414] attn_weights_entropy = tensor([2.0649, 1.6505, 4.0800, 3.8988, 3.9109, 4.0868, 3.2733, 4.1840],
       device='cuda:0'), covar=tensor([0.1251, 0.1291, 0.0081, 0.0142, 0.0145, 0.0086, 0.0117, 0.0077],
       device='cuda:0'), in_proj_covar=tensor([0.0139, 0.0150, 0.0113, 0.0156, 0.0131, 0.0125, 0.0105, 0.0106],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
       device='cuda:0')
2022-12-08 23:53:50,687 INFO [decode.py:463] batch 100/?, cuts processed until now is 13420
2022-12-08 23:53:55,847 INFO [zipformer.py:1414] attn_weights_entropy = tensor([4.9753, 2.8328, 4.9959, 2.8165, 4.7903, 2.2568, 3.6416, 4.6590],
       device='cuda:0'), covar=tensor([0.0528, 0.5202, 0.0432, 1.3318, 0.0421, 0.5119, 0.1614, 0.0291],
       device='cuda:0'), in_proj_covar=tensor([0.0221, 0.0205, 0.0174, 0.0283, 0.0196, 0.0207, 0.0197, 0.0178],
       device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004],
       device='cuda:0')
2022-12-08 23:53:58,036 INFO [decode.py:479] The transcripts are stored in pruned_transducer_stateless7/exp/v1/greedy_search/recogs-test_ihm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:53:58,352 INFO [utils.py:536] [test_ihm-greedy_search] %WER 12.21% [25615 / 209845, 2007 ins, 7895 del, 15713 sub ]
2022-12-08 23:53:58,963 INFO [decode.py:492] Wrote detailed error stats to pruned_transducer_stateless7/exp/v1/greedy_search/errs-test_ihm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:53:58,964 INFO [decode.py:508] 
For test_ihm, WER of different settings are:
greedy_search	12.21	best for test_ihm

2022-12-08 23:53:58,964 INFO [decode.py:687] Decoding eval_sdm
2022-12-08 23:54:00,431 INFO [decode.py:463] batch 0/?, cuts processed until now is 58
2022-12-08 23:54:09,011 INFO [zipformer.py:1414] attn_weights_entropy = tensor([2.9497, 2.8796, 3.7148, 2.5176, 2.4075, 2.9990, 1.6958, 2.9434],
       device='cuda:0'), covar=tensor([0.1004, 0.0976, 0.0427, 0.2647, 0.2157, 0.0924, 0.3887, 0.0774],
       device='cuda:0'), in_proj_covar=tensor([0.0070, 0.0085, 0.0077, 0.0085, 0.0104, 0.0072, 0.0116, 0.0077],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003],
       device='cuda:0')
2022-12-08 23:54:21,106 INFO [zipformer.py:1414] attn_weights_entropy = tensor([4.0325, 3.8665, 3.9421, 3.9902, 3.6039, 3.2852, 4.1056, 3.8995],
       device='cuda:0'), covar=tensor([0.0388, 0.0286, 0.0419, 0.0433, 0.0408, 0.0515, 0.0381, 0.0510],
       device='cuda:0'), in_proj_covar=tensor([0.0122, 0.0118, 0.0125, 0.0134, 0.0129, 0.0102, 0.0145, 0.0125],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
       device='cuda:0')
2022-12-08 23:54:23,976 INFO [decode.py:479] The transcripts are stored in pruned_transducer_stateless7/exp/v1/greedy_search/recogs-eval_sdm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:54:24,077 INFO [utils.py:536] [eval_sdm-greedy_search] %WER 23.70% [19222 / 81111, 1683 ins, 6073 del, 11466 sub ]
2022-12-08 23:54:24,332 INFO [decode.py:492] Wrote detailed error stats to pruned_transducer_stateless7/exp/v1/greedy_search/errs-eval_sdm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:54:24,333 INFO [decode.py:508] 
For eval_sdm, WER of different settings are:
greedy_search	23.7	best for eval_sdm

2022-12-08 23:54:24,333 INFO [decode.py:687] Decoding test_sdm
2022-12-08 23:54:26,054 INFO [decode.py:463] batch 0/?, cuts processed until now is 49
2022-12-08 23:54:27,800 INFO [zipformer.py:1414] attn_weights_entropy = tensor([5.8932, 5.8649, 5.7222, 5.8181, 5.3845, 5.4143, 5.9574, 5.6303],
       device='cuda:0'), covar=tensor([0.0422, 0.0200, 0.0364, 0.0455, 0.0434, 0.0144, 0.0327, 0.0613],
       device='cuda:0'), in_proj_covar=tensor([0.0122, 0.0118, 0.0125, 0.0134, 0.0129, 0.0102, 0.0145, 0.0125],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
       device='cuda:0')
2022-12-08 23:54:31,870 INFO [zipformer.py:1414] attn_weights_entropy = tensor([5.0367, 4.5032, 4.6101, 4.9917, 4.5509, 4.6262, 4.8931, 4.4161],
       device='cuda:0'), covar=tensor([0.0253, 0.1510, 0.0270, 0.0302, 0.0832, 0.0292, 0.0559, 0.0433],
       device='cuda:0'), in_proj_covar=tensor([0.0149, 0.0248, 0.0167, 0.0163, 0.0160, 0.0127, 0.0252, 0.0145],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
       device='cuda:0')
2022-12-08 23:54:36,104 INFO [zipformer.py:1414] attn_weights_entropy = tensor([3.5049, 3.0396, 3.1109, 2.3003, 2.9128, 3.3196, 3.3253, 2.8653],
       device='cuda:0'), covar=tensor([0.0876, 0.2228, 0.1388, 0.2084, 0.1478, 0.0834, 0.1264, 0.1702],
       device='cuda:0'), in_proj_covar=tensor([0.0124, 0.0170, 0.0124, 0.0117, 0.0121, 0.0128, 0.0106, 0.0128],
       device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005],
       device='cuda:0')
2022-12-08 23:54:47,951 INFO [zipformer.py:1414] attn_weights_entropy = tensor([2.9691, 2.6751, 2.9312, 3.1152, 2.8372, 2.3588, 2.9794, 3.0298],
       device='cuda:0'), covar=tensor([0.0107, 0.0177, 0.0195, 0.0116, 0.0141, 0.0331, 0.0139, 0.0184],
       device='cuda:0'), in_proj_covar=tensor([0.0258, 0.0233, 0.0347, 0.0293, 0.0235, 0.0280, 0.0265, 0.0260],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002],
       device='cuda:0')
2022-12-08 23:55:21,831 INFO [decode.py:463] batch 100/?, cuts processed until now is 13420
2022-12-08 23:55:29,332 INFO [decode.py:479] The transcripts are stored in pruned_transducer_stateless7/exp/v1/greedy_search/recogs-test_sdm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:55:29,621 INFO [utils.py:536] [test_sdm-greedy_search] %WER 26.41% [55414 / 209845, 4503 ins, 19379 del, 31532 sub ]
2022-12-08 23:55:30,282 INFO [decode.py:492] Wrote detailed error stats to pruned_transducer_stateless7/exp/v1/greedy_search/errs-test_sdm-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:55:30,283 INFO [decode.py:508] 
For test_sdm, WER of different settings are:
greedy_search	26.41	best for test_sdm

2022-12-08 23:55:30,283 INFO [decode.py:687] Decoding eval_gss
2022-12-08 23:55:31,773 INFO [decode.py:463] batch 0/?, cuts processed until now is 58
2022-12-08 23:55:34,158 INFO [zipformer.py:1414] attn_weights_entropy = tensor([3.2471, 3.7230, 3.3454, 2.8730, 2.7644, 3.7741, 3.4475, 1.8443],
       device='cuda:0'), covar=tensor([0.3240, 0.0695, 0.1965, 0.1750, 0.1144, 0.0470, 0.1355, 0.3201],
       device='cuda:0'), in_proj_covar=tensor([0.0138, 0.0066, 0.0052, 0.0054, 0.0082, 0.0064, 0.0085, 0.0091],
       device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005],
       device='cuda:0')
2022-12-08 23:55:37,669 INFO [zipformer.py:1414] attn_weights_entropy = tensor([3.7584, 3.8680, 4.3876, 3.2812, 2.7117, 3.5261, 2.1305, 3.5684],
       device='cuda:0'), covar=tensor([0.0716, 0.0548, 0.0428, 0.2018, 0.2384, 0.0745, 0.3972, 0.0912],
       device='cuda:0'), in_proj_covar=tensor([0.0070, 0.0085, 0.0077, 0.0085, 0.0104, 0.0072, 0.0116, 0.0077],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003],
       device='cuda:0')
2022-12-08 23:55:41,813 INFO [zipformer.py:1414] attn_weights_entropy = tensor([4.2555, 3.7707, 3.0407, 4.4182, 4.2751, 4.2557, 3.6943, 2.9354],
       device='cuda:0'), covar=tensor([0.0750, 0.1279, 0.4109, 0.0665, 0.0696, 0.1311, 0.1328, 0.4549],
       device='cuda:0'), in_proj_covar=tensor([0.0237, 0.0272, 0.0252, 0.0220, 0.0279, 0.0268, 0.0232, 0.0237],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003],
       device='cuda:0')
2022-12-08 23:55:55,409 INFO [decode.py:479] The transcripts are stored in pruned_transducer_stateless7/exp/v1/greedy_search/recogs-eval_gss-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:55:55,506 INFO [utils.py:536] [eval_gss-greedy_search] %WER 12.24% [9930 / 81111, 915 ins, 2606 del, 6409 sub ]
2022-12-08 23:55:55,743 INFO [decode.py:492] Wrote detailed error stats to pruned_transducer_stateless7/exp/v1/greedy_search/errs-eval_gss-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:55:55,744 INFO [decode.py:508] 
For eval_gss, WER of different settings are:
greedy_search	12.24	best for eval_gss

2022-12-08 23:55:55,744 INFO [decode.py:687] Decoding test_gss
2022-12-08 23:55:57,430 INFO [decode.py:463] batch 0/?, cuts processed until now is 49
2022-12-08 23:56:44,408 INFO [zipformer.py:1414] attn_weights_entropy = tensor([2.0077, 1.3975, 3.4059, 2.9515, 3.0863, 3.3663, 2.8472, 3.3858],
       device='cuda:0'), covar=tensor([0.0407, 0.0664, 0.0061, 0.0240, 0.0239, 0.0078, 0.0193, 0.0094],
       device='cuda:0'), in_proj_covar=tensor([0.0139, 0.0150, 0.0113, 0.0156, 0.0131, 0.0125, 0.0105, 0.0106],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
       device='cuda:0')
2022-12-08 23:56:53,454 INFO [decode.py:463] batch 100/?, cuts processed until now is 13420
2022-12-08 23:57:00,993 INFO [decode.py:479] The transcripts are stored in pruned_transducer_stateless7/exp/v1/greedy_search/recogs-test_gss-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:57:01,279 INFO [utils.py:536] [test_gss-greedy_search] %WER 14.99% [31450 / 209845, 2293 ins, 9720 del, 19437 sub ]
2022-12-08 23:57:01,910 INFO [decode.py:492] Wrote detailed error stats to pruned_transducer_stateless7/exp/v1/greedy_search/errs-test_gss-greedy_search-epoch-15-avg-8-context-2-max-sym-per-frame-1.txt
2022-12-08 23:57:01,911 INFO [decode.py:508] 
For test_gss, WER of different settings are:
greedy_search	14.99	best for test_gss

2022-12-08 23:57:01,912 INFO [decode.py:703] Done!