#!/bin/bash filter_regexp=. [ $# -ge 1 ] && filter_regexp=$1 for x in exp/*/decode*; do [ -d $x ] && grep WER $x/wer_* | utils/best_wer.sh; done 2>/dev/null for x in exp/{mono,tri,sgmm,nnet,dnn,lstm}*/decode*; do [ -d $x ] && grep Sum $x/score_*/*.sys | utils/best_wer.sh; done 2>/dev/null | grep $filter_regexp exit 0 # GMM results [these are up to date] %WER 30.3 | 507 17783 | 73.5 19.5 7.0 3.9 30.3 96.3 | 0.006 | exp/tri1/decode_nosp_dev/score_12_0.0/ctm.filt.filt.sys %WER 30.5 | 1155 27500 | 73.3 20.9 5.8 3.8 30.5 93.9 | 0.029 | exp/tri1/decode_nosp_test/score_11_0.0/ctm.filt.filt.sys %WER 25.4 | 507 17783 | 78.4 15.8 5.9 3.8 25.4 95.3 | -0.075 | exp/tri2/decode_dev/score_14_0.5/ctm.filt.filt.sys %WER 26.2 | 507 17783 | 78.0 16.4 5.6 4.2 26.2 95.3 | -0.066 | exp/tri2/decode_nosp_dev/score_13_0.0/ctm.filt.filt.sys %WER 25.2 | 1155 27500 | 78.1 17.0 5.0 3.3 25.2 91.9 | -0.005 | exp/tri2/decode_nosp_test/score_13_0.0/ctm.filt.filt.sys %WER 24.5 | 1155 27500 | 78.7 16.5 4.8 3.2 24.5 91.4 | -0.014 | exp/tri2/decode_test/score_13_0.5/ctm.filt.filt.sys %WER 21.6 | 507 17783 | 82.4 12.9 4.6 4.0 21.6 92.9 | -0.094 | exp/tri3/decode_dev/score_12_0.5/ctm.filt.filt.sys %WER 19.7 | 1155 27500 | 83.6 13.1 3.3 3.3 19.7 89.4 | -0.037 | exp/tri3/decode_test/score_14_0.0/ctm.filt.filt.sys %WER 18.5 | 507 17783 | 84.3 11.2 4.5 2.8 18.5 90.5 | -0.104 | exp/tri3_mmi_b0.1/decode_dev_it4/score_10_1.0/ctm.filt.filt.sys %WER 17.1 | 1155 27500 | 85.2 11.3 3.5 2.3 17.1 86.7 | 0.049 | exp/tri3_mmi_b0.1/decode_test_it4/score_11_0.5/ctm.filt.filt.sys # ========= Most of the results below this line are little out of date; we made some changes # to the training segmentation (add a little padding), and to the text preparation # (merge all suffixes like 's and 're instead of leaving them separate), and # results below this point were not re-run. # Karel's DNN %WER 15.0 | 1155 27512 | 87.2 9.7 3.0 2.2 15.0 83.6 | -0.014 | exp/dnn4_pretrain-dbn_dnn/decode_test/score_10/ctm.filt.filt.sys %WER 13.3 | 1155 27512 | 88.9 8.8 2.3 2.2 13.3 81.4 | -0.063 | exp/dnn4_pretrain-dbn_dnn_smbr_i1lats/decode_test_it4/score_11/ctm.filt.filt.sys %WER 12.8 | 1155 27512 | 89.0 8.4 2.6 1.8 12.8 80.2 | -0.122 | exp/dnn8e_BN_pretrain-dbn_dnn/decode_test/score_10/ctm.filt.filt.sys %WER 11.7 | 1155 27512 | 90.2 7.7 2.1 2.0 11.7 78.7 | -0.110 | exp/dnn8f_BN_pretrain-dbn_dnn_smbr/decode_test_it4/score_12/ctm.filt.filt.sys # Karel's DNN + rescore %WER 13.5 | 1155 27512 | 88.4 8.5 3.1 1.9 13.5 81.6 | -0.074 | exp/dnn4_pretrain-dbn_dnn/decode_test.rescore/score_10/ctm.filt.filt.sys %WER 11.9 | 1155 27512 | 90.0 7.6 2.3 1.9 11.9 78.2 | -0.110 | exp/dnn4_pretrain-dbn_dnn_smbr_i1lats/decode_test_it4.rescore/score_11/ctm.filt.filt.sys %WER 10.6 | 1155 27512 | 91.0 6.7 2.3 1.6 10.6 74.4 | -0.194 | exp/dnn8f_BN_pretrain-dbn_dnn_smbr/decode_test_it4.rescore/score_12/ctm.filt.filt.sys # multi-splice + i-vector + perturbed %WER 14.0 | 507 17792 | 88.5 8.4 3.0 2.5 14.0 88.0 | -0.074 | exp/nnet2_online/nnet_ms_sp/decode_dev/score_12/ctm.filt.filt.sys %WER 13.3 | 1155 27512 | 88.7 8.7 2.6 2.0 13.3 81.5 | -0.097 | exp/nnet2_online/nnet_ms_sp/decode_test/score_10/ctm.filt.filt.sys %WER 13.2 | 1155 27512 | 88.6 8.5 2.8 1.9 13.2 81.6 | -0.102 | exp/nnet2_online/nnet_ms_sp_online/decode_test/score_11/ctm.filt.filt.sys %WER 13.6 | 1155 27512 | 88.5 8.9 2.6 2.1 13.6 82.7 | -0.095 | exp/nnet2_online/nnet_ms_sp_online/decode_test_utt/score_10/ctm.filt.filt.sys # multi-splice + i-vector + perturbed + rescore %WER 11.9 | 1155 27512 | 89.9 7.5 2.6 1.8 11.9 77.9 | -0.177 | exp/nnet2_online/nnet_ms_sp/decode_test.rescore/score_10/ctm.filt.filt.sys %WER 11.8 | 1155 27512 | 90.0 7.4 2.6 1.8 11.8 77.6 | -0.300 | exp/nnet2_online/nnet_ms_sp_online/decode_test.rescore/score_10/ctm.filt.filt.sys %WER 11.8 | 1155 27512 | 89.9 7.4 2.7 1.8 11.8 79.0 | -0.233 | exp/nnet2_online/nnet_ms_sp_online/decode_test_utt_offline.rescore/score_11/ctm.filt.filt.sys %WER 12.3 | 1155 27512 | 89.5 7.6 3.0 1.8 12.3 80.5 | -0.200 | exp/nnet2_online/nnet_ms_sp_online/decode_test_utt.rescore/score_12/ctm.filt.filt.sys # multi-splice + i-vector + perturbed + sMBR training %WER 13.2 | 1155 27512 | 88.6 8.5 2.8 1.9 13.2 81.6 | -0.102 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch0_test/score_11/ctm.filt.filt.sys %WER 12.8 | 1155 27512 | 89.1 8.4 2.5 2.0 12.8 81.1 | -0.108 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch1_test/score_10/ctm.filt.filt.sys %WER 12.6 | 1155 27512 | 89.3 8.2 2.6 1.9 12.6 81.1 | -0.062 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch2_test/score_11/ctm.filt.filt.sys %WER 12.5 | 1155 27512 | 89.4 8.1 2.4 1.9 12.5 81.3 | -0.064 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch3_test/score_11/ctm.filt.filt.sys %WER 12.5 | 1155 27512 | 89.6 8.1 2.2 2.1 12.5 80.8 | -0.107 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch4_test/score_10/ctm.filt.filt.sys # multi-splice + i-vector + perturbed + sMBR training + rescore %WER 11.8 | 1155 27512 | 90.0 7.4 2.6 1.8 11.8 77.6 | -0.300 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch0_test.rescore/score_10/ctm.filt.filt.sys %WER 11.4 | 1155 27512 | 90.4 7.1 2.5 1.7 11.4 76.9 | -0.253 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch1_test.rescore/score_10/ctm.filt.filt.sys %WER 11.2 | 1155 27512 | 90.6 7.0 2.4 1.7 11.2 76.5 | -0.240 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch2_test.rescore/score_10/ctm.filt.filt.sys %WER 11.0 | 1155 27512 | 90.7 6.8 2.5 1.7 11.0 75.8 | -0.232 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch3_test.rescore/score_11/ctm.filt.filt.sys %WER 10.9 | 1155 27512 | 90.7 6.8 2.4 1.7 10.9 75.8 | -0.230 | exp/nnet2_online/nnet_ms_sp_smbr_0.000005/decode_epoch4_test.rescore/score_11/ctm.filt.filt.sys #---------------------------------(Old LM) Provided for reference----------------------------------------------- # DEV SPEAKERS: %WER 36.0 | 507 17792 | 69.7 23.3 7.0 5.7 36.0 98.2 | 0.009 | exp/tri1/decode_dev/score_12/ctm.filt.filt.sys %WER 32.1 | 507 17792 | 73.8 20.1 6.1 6.0 32.1 97.2 | -0.092 | exp/tri2/decode_dev/score_15/ctm.filt.filt.sys %WER 27.4 | 507 17792 | 78.3 16.7 5.0 5.7 27.4 96.4 | -0.232 | exp/tri3/decode_dev/score_16/ctm.filt.filt.sys %WER 24.0 | 507 17792 | 80.6 14.7 4.7 4.6 24.0 94.1 | -0.098 | exp/tri3_mmi_b0.1/decode_dev_it4/score_15/ctm.filt.filt.sys # Karel's DNN %WER 21.7 | 507 17792 | 83.3 13.1 3.7 5.0 21.7 93.5 | -0.175 | exp/dnn4_pretrain-dbn_dnn/decode_dev/score_11/ctm.filt.filt.sys %WER 20.0 | 507 17792 | 84.7 12.1 3.1 4.7 20.0 92.5 | -0.202 | exp/dnn4_pretrain-dbn_dnn_smbr/decode_dev/score_12/ctm.filt.filt.sys %WER 19.4 | 507 17792 | 85.2 12.0 2.8 4.6 19.4 91.9 | -0.223 | exp/dnn4_pretrain-dbn_dnn_smbr_i1lats/decode_dev/score_13/ctm.filt.filt.sys # TEST SPEAKERS: # GMMs %WER 34.7 | 1155 27512 | 69.9 23.7 6.4 4.6 34.7 97.3 | 0.080 | exp/tri1/decode_test/score_13/ctm.filt.filt.sys %WER 29.8 | 1155 27512 | 74.8 20.2 5.0 4.6 29.8 95.9 | -0.015 | exp/tri2/decode_test/score_14/ctm.filt.filt.sys %WER 24.6 | 1155 27512 | 79.6 16.3 4.1 4.2 24.6 93.6 | -0.050 | exp/tri3/decode_test/score_16/ctm.filt.filt.sys %WER 21.6 | 1155 27512 | 82.3 14.2 3.5 3.9 21.6 91.9 | 0.043 | exp/tri3_mmi_b0.1/decode_test_it4/score_13/ctm.filt.filt.sys # Karel's DNN %WER 19.1 | 1155 27512 | 84.4 12.1 3.5 3.5 19.1 90.0 | -0.025 | exp/dnn4_pretrain-dbn_dnn/decode_test/score_12/ctm.filt.filt.sys %WER 17.7 | 1155 27512 | 85.8 11.3 2.9 3.5 17.7 88.7 | -0.049 | exp/dnn4_pretrain-dbn_dnn_smbr/decode_test/score_13/ctm.filt.filt.sys %WER 17.2 | 1155 27512 | 86.2 11.0 2.7 3.4 17.2 87.9 | -0.063 | exp/dnn4_pretrain-dbn_dnn_smbr_i1lats/decode_test/score_14/ctm.filt.filt.sys # multi-splice + i-vector %WER 17.9 | 1155 27512 | 85.6 11.2 3.1 3.6 17.9 88.1 | -0.105 | exp/nnet2_online/nnet_ms_a/decode_test/score_12/ctm.filt.filt.sys %WER 17.9 | 1155 27512 | 85.6 11.2 3.1 3.6 17.9 88.1 | -0.187 | exp/nnet2_online/nnet_ms_a_online/decode_test/score_12/ctm.filt.filt.sys %WER 18.0 | 1155 27512 | 85.3 11.3 3.4 3.3 18.0 87.9 | -0.196 | exp/nnet2_online/nnet_ms_a_online/decode_test_utt/score_13/ctm.filt.filt.sys %WER 17.7 | 1155 27512 | 85.6 11.1 3.3 3.3 17.7 87.4 | -0.177 | exp/nnet2_online/nnet_ms_a_online/decode_test_utt_offline/score_13/ctm.filt.filt.sys %WER 17.9 | 1155 27512 | 85.6 11.2 3.1 3.6 17.9 88.1 | -0.187 | exp/nnet2_online/nnet_ms_a_smbr_0.000005/decode_epoch0_test/score_12/ctm.filt.filt.sys %WER 17.6 | 1155 27512 | 86.2 11.0 2.7 3.9 17.6 87.7 | -0.191 | exp/nnet2_online/nnet_ms_a_smbr_0.000005/decode_epoch1_test/score_12/ctm.filt.filt.sys %WER 17.3 | 1155 27512 | 86.4 10.9 2.7 3.7 17.3 87.8 | -0.191 | exp/nnet2_online/nnet_ms_a_smbr_0.000005/decode_epoch2_test/score_13/ctm.filt.filt.sys %WER 17.2 | 1155 27512 | 86.5 10.9 2.6 3.7 17.2 87.6 | -0.190 | exp/nnet2_online/nnet_ms_a_smbr_0.000005/decode_epoch3_test/score_13/ctm.filt.filt.sys %WER 17.2 | 1155 27512 | 86.6 10.8 2.6 3.8 17.2 87.6 | -0.193 | exp/nnet2_online/nnet_ms_a_smbr_0.000005/decode_epoch4_test/score_13/ctm.filt.filt.sys # multi-splice + i-vector + perturbed %WER 17.2 | 1155 27512 | 86.3 10.8 2.9 3.5 17.2 87.1 | -0.126 | exp/nnet2_online/nnet_ms_sp/decode_test/score_12/ctm.filt.filt.sys %WER 17.2 | 1155 27512 | 86.3 10.8 3.0 3.5 17.2 87.0 | -0.229 | exp/nnet2_online/nnet_ms_sp_online/decode_test/score_12/ctm.filt.filt.sys %WER 17.6 | 1155 27512 | 85.9 11.1 3.0 3.5 17.6 87.8 | -0.210 | exp/nnet2_online/nnet_ms_sp_online/decode_test_utt/score_12/ctm.filt.filt.sys %WER 17.2 | 1155 27512 | 86.5 10.8 2.7 3.7 17.2 87.4 | -0.236 | exp/nnet2_online/nnet_ms_sp_online/decode_test_utt_offline/score_11/ctm.filt.filt.sys # new dict, lm, %WER 13.3 | 1155 27512 | 88.7 8.7 2.6 2.0 13.3 81.5 | -0.097 | exp/nnet2_online/nnet_ms_sp/decode_test/score_10/ctm.filt.filt.sys %WER 13.2 | 1155 27512 | 88.6 8.5 2.8 1.9 13.2 81.6 | -0.102 | exp/nnet2_online/nnet_ms_sp_online/decode_test/score_11/ctm.filt.filt.sys %WER 13.6 | 1155 27512 | 88.5 8.9 2.6 2.1 13.6 82.7 | -0.095 | exp/nnet2_online/nnet_ms_sp_online/decode_test_utt/score_10/ctm.filt.filt.sys ### LSTM vs. DNN ### # DNN on MFCC-fMLLR %WER 19.1 | 1155 27512 | 84.4 12.1 3.5 3.5 19.1 90.0 | -0.025 | exp/dnn4_pretrain-dbn_dnn/decode_test/score_12/ctm.filt.filt.sys # DNN on FBANK-pitch (we see pitch compensated degradation of not having fMLLR), %WER 19.2 | 1155 27512 | 84.4 12.3 3.3 3.6 19.2 89.2 | -0.021 | exp/dnn4d-fbank_pretrain-dbn_dnn/decode_test/score_12/ctm.filt.filt.sys # LSTM %WER 20.3 | 1155 27512 | 83.3 13.2 3.5 3.6 20.3 90.7 | -0.176 | exp/lstm4f_ClipGradient5_lrate1e-4/decode_test/score_11/ctm.filt.filt.sys # 2xLSTM TODO...