Yannick Estève / ONTRAC-Kaldi

Blame view

egs/chime4/s5_6ch/RESULTS 7.91 KB
edit raw normal view history
8dcb6dfcb   Yannick Estève   first commit 
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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
 
  # CHiME-4 6ch track results
  # The result is based on Hori et al, "The MERL/SRI system for the 3rd CHiME challenge using beamforming,
  # robust feature extraction, and advanced speech recognition," in Proc. ASRU'15,
  # and please refer the paper if you think the baseline useful.
  # Note that the following result is different from that in the paper since we don't include
  # SRI's robust features and system combination
  
  GMM noisy multi-condition with beamformit
  exp/tri3b_tr05_multi_noisy/best_wer_beamformit_5mics.result
  -------------------
  best overall dt05 WER 13.67% (language model weight = 11)
  -------------------
  dt05_simu WER: 14.30% (Average), 12.80% (BUS), 17.05% (CAFE), 11.90% (PEDESTRIAN), 15.46% (STREET)
  -------------------
  dt05_real WER: 13.03% (Average), 16.03% (BUS), 12.80% (CAFE), 10.02% (PEDESTRIAN), 13.27% (STREET)
  -------------------
  et05_simu WER: 21.30% (Average), 15.73% (BUS), 22.94% (CAFE), 22.51% (PEDESTRIAN), 24.04% (STREET)
  -------------------
  et05_real WER: 21.83% (Average), 30.17% (BUS), 20.66% (CAFE), 19.82% (PEDESTRIAN), 16.68% (STREET)
  -------------------
  
  GMM noisy multi-condition with blstm_gev
  exp/tri3b_tr05_multi_noisy/best_wer_blstm_gev.result
  -------------------
  best overall dt05 WER 11.17% (language model weight = 12)
  -------------------
  dt05_simu WER: 11.44% (Average), 9.78% (BUS), 14.37% (CAFE), 10.10% (PEDESTRIAN), 11.50% (STREET)
  -------------------
  dt05_real WER: 10.91% (Average), 11.21% (BUS), 11.24% (CAFE), 10.34% (PEDESTRIAN), 10.84% (STREET)
  -------------------
  et05_simu WER: 13.54% (Average), 11.65% (BUS), 14.90% (CAFE), 13.73% (PEDESTRIAN), 13.86% (STREET)
  -------------------
  et05_real WER: 14.62% (Average), 16.43% (BUS), 15.43% (CAFE), 12.99% (PEDESTRIAN), 13.63% (STREET)
  -------------------
  
  DNN sMBR with beamformit
  exp/tri4a_dnn_tr05_multi_noisy_smbr_i1lats/best_wer_beamformit_5mics.result
  -------------------
  best overall dt05 WER 8.60% (language model weight = 11)
   (Number of iterations = 4)
  -------------------
  dt05_simu WER: 9.07% (Average), 8.44% (BUS), 10.63% (CAFE), 7.39% (PEDESTRIAN), 9.82% (STREET)
  -------------------
  dt05_real WER: 8.14% (Average), 10.22% (BUS), 8.19% (CAFE), 5.69% (PEDESTRIAN), 8.45% (STREET)
  -------------------
  et05_simu WER: 14.23% (Average), 10.72% (BUS), 15.52% (CAFE), 13.90% (PEDESTRIAN), 16.77% (STREET)
  -------------------
  et05_real WER: 15.00% (Average), 21.74% (BUS), 13.58% (CAFE), 12.84% (PEDESTRIAN), 11.86% (STREET)
  -------------------
  
  DNN sMBR with blstm_gev
  exp/tri4a_dnn_tr05_multi_noisy_smbr_i1lats/best_wer_blstm_gev.result
  -------------------
  best overall dt05 WER 7.38% (language model weight = 11)
   (Number of iterations = 4)
  -------------------
  dt05_simu WER: 7.49% (Average), 5.93% (BUS), 9.69% (CAFE), 6.73% (PEDESTRIAN), 7.61% (STREET)
  -------------------
  dt05_real WER: 7.28% (Average), 7.83% (BUS), 7.80% (CAFE), 6.37% (PEDESTRIAN), 7.11% (STREET)
  -------------------
  et05_simu WER: 9.54% (Average), 8.18% (BUS), 10.87% (CAFE), 9.81% (PEDESTRIAN), 9.32% (STREET)
  -------------------
  et05_real WER: 9.77% (Average), 11.42% (BUS), 10.22% (CAFE), 9.23% (PEDESTRIAN), 8.22% (STREET)
  -------------------
  
  RNNLM with beamformit
  exp/tri4a_dnn_tr05_multi_noisy_smbr_lmrescore/best_wer_beamformit_5mics_rnnlm_5k_h300_w0.5_n100.result
  -------------------
  best overall dt05 WER 6.27% (language model weight = 12)
  -------------------
  dt05_simu WER: 6.77% (Average), 6.02% (BUS), 8.10% (CAFE), 5.49% (PEDESTRIAN), 7.48% (STREET)
  -------------------
  dt05_real WER: 5.76% (Average), 7.39% (BUS), 5.77% (CAFE), 3.72% (PEDESTRIAN), 6.18% (STREET)
  -------------------
  et05_simu WER: 10.90% (Average), 7.68% (BUS), 11.54% (CAFE), 10.31% (PEDESTRIAN), 14.06% (STREET)
  -------------------
  et05_real WER: 11.51% (Average), 16.86% (BUS), 10.18% (CAFE), 9.83% (PEDESTRIAN), 9.19% (STREET)
  -------------------
  
  ######## Advanced baseline
  ######## All 6 channel training, enhanced data training, Lattice-free MMI TDNN, BLSTM-mask-based GEV beamformer
  
  TDNN with beamformit
  exp/chain/tdnn1d_sp/best_wer_beamformit_5mics.result
  -------------------
  best overall dt05 WER 6.04% (language model weight = 9)
  -------------------
  dt05_simu WER: 6.25% (Average), 5.71% (BUS), 6.92% (CAFE), 5.37% (PEDESTRIAN), 7.02% (STREET)
  -------------------
  dt05_real WER: 5.83% (Average), 7.48% (BUS), 5.28% (CAFE), 4.43% (PEDESTRIAN), 6.13% (STREET)
  -------------------
  et05_simu WER: 10.30% (Average), 7.34% (BUS), 10.37% (CAFE), 10.05% (PEDESTRIAN), 13.43% (STREET)
  -------------------
  et05_real WER: 9.67% (Average), 12.71% (BUS), 8.33% (CAFE), 8.20% (PEDESTRIAN), 9.45% (STREET)
  -------------------
  
  TDNN+RNNLM with beamformit
  exp/chain/tdnn1d_sp_smbr_lmrescore/best_wer_beamformit_5mics_rnnlm_5k_h300_w0.5_n100.result
  -------------------
  best overall dt05 WER 4.15% (language model weight = 9)
  -------------------
  dt05_simu WER: 4.33% (Average), 3.95% (BUS), 4.87% (CAFE), 3.53% (PEDESTRIAN), 4.97% (STREET)
  -------------------
  dt05_real WER: 3.97% (Average), 5.38% (BUS), 3.19% (CAFE), 2.94% (PEDESTRIAN), 4.37% (STREET)
  -------------------
  et05_simu WER: 7.39% (Average), 4.87% (BUS), 7.58% (CAFE), 7.15% (PEDESTRIAN), 9.96% (STREET)
  -------------------
  et05_real WER: 7.04% (Average), 9.89% (BUS), 5.49% (CAFE), 5.70% (PEDESTRIAN), 7.10% (STREET)
  -------------------
  
  TDNN using 6 channel data plus enhanced data with beamformit
  exp/chain/tdnn7a_sp/best_wer_beamformit_5mics.result
  -------------------
  best overall dt05 WER 5.80% (language model weight = 10)
  -------------------
  dt05_simu WER: 6.19% (Average), 5.96% (BUS), 6.78% (CAFE), 5.10% (PEDESTRIAN), 6.92% (STREET)
  -------------------
  dt05_real WER: 5.41% (Average), 6.86% (BUS), 4.87% (CAFE), 4.00% (PEDESTRIAN), 5.91% (STREET)
  -------------------
  et05_simu WER: 10.26% (Average), 7.68% (BUS), 10.40% (CAFE), 10.16% (PEDESTRIAN), 12.79% (STREET)
  -------------------
  et05_real WER: 9.63% (Average), 13.46% (BUS), 7.98% (CAFE), 8.13% (PEDESTRIAN), 8.97% (STREET)
  -------------------
  
  TDNN+RNNLM using 6 channel data plus enhanced data with beamformit
  exp/chain/tdnn7a_sp_smbr_lmrescore/best_wer_beamformit_5mics_rnnlm_5k_h300_w0.5_n100.result
  compute dt05 WER for each location
  -------------------
  best overall dt05 WER 4.02% (language model weight = 11)
  -------------------
  dt05_simu WER: 4.31% (Average), 4.04% (BUS), 4.88% (CAFE), 3.38% (PEDESTRIAN), 4.94% (STREET)
  -------------------
  dt05_real WER: 3.74% (Average), 4.62% (BUS), 3.17% (CAFE), 3.02% (PEDESTRIAN), 4.14% (STREET)
  -------------------
  et05_simu WER: 7.49% (Average), 5.16% (BUS), 7.21% (CAFE), 7.45% (PEDESTRIAN), 10.14% (STREET)
  -------------------
  et05_real WER: 6.84% (Average), 9.74% (BUS), 5.38% (CAFE), 5.25% (PEDESTRIAN), 7.00% (STREET)
  -------------------
  
  TDNN+RNNLM using 6 channel data plus enhanced data with blstm_gev
  exp/chain/tdnn1a_sp_smbr_lmrescore/best_wer_blstm_gev_rnnlm_5k_h300_w0.5_n100.result
  -------------------
  best overall dt05 WER 3.01% (language model weight = 10)
  -------------------
  dt05_simu WER: 3.10% (Average), 2.60% (BUS), 4.07% (CAFE), 2.80% (PEDESTRIAN), 2.92% (STREET)
  -------------------
  dt05_real WER: 2.93% (Average), 3.32% (BUS), 2.83% (CAFE), 2.63% (PEDESTRIAN), 2.93% (STREET)
  -------------------
  et05_simu WER: 3.95% (Average), 3.29% (BUS), 4.71% (CAFE), 4.30% (PEDESTRIAN), 3.53% (STREET)
  -------------------
  et05_real WER: 4.04% (Average), 4.94% (BUS), 3.66% (CAFE), 3.66% (PEDESTRIAN), 3.90% (STREET)
  -------------------
  
  TDNN+LSTMLM using 6 channel data plus enhanced data with blstm_gev
  exp/chain/tdnn1a_sp_smbr_lmrescore/best_wer_blstm_gev_rnnlm_lstm_1a_w0.5_n100.result
  -------------------
  best overall dt05 WER 2.00% (language model weight = 11)
  -------------------
  dt05_simu WER: 2.10% (Average), 2.06% (BUS), 2.58% (CAFE), 1.73% (PEDESTRIAN), 2.02% (STREET)
  -------------------
  dt05_real WER: 1.90% (Average), 2.05% (BUS), 1.78% (CAFE), 1.68% (PEDESTRIAN), 2.09% (STREET)
  -------------------
  et05_simu WER: 2.66% (Average), 2.33% (BUS), 2.73% (CAFE), 2.93% (PEDESTRIAN), 2.63% (STREET)
  -------------------
  et05_real WER: 2.74% (Average), 3.05% (BUS), 2.45% (CAFE), 2.65% (PEDESTRIAN), 2.82% (STREET)
  -------------------