Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Make a more detailed description for tdnn_lstm_1a in Fisher_SWBD #2122

Merged
merged 1 commit into from
Jan 3, 2018
Merged

Make a more detailed description for tdnn_lstm_1a in Fisher_SWBD #2122

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions egs/fisher_swbd/s5/local/chain/run_blstm_6j.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,6 @@
#!/bin/bash
# Copyright 2017 University of Chinese Academy of Sciences (UCAS) Gaofeng Cheng
# Apache 2.0

# The model training procedure is similar to run_blstm_6j.sh under egs/swbd/s5c

Expand Down
11 changes: 8 additions & 3 deletions egs/fisher_swbd/s5/local/chain/run_tdnn_lstm_1a.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,10 @@
#!/bin/bash
# same as run_tdnn_opgru_1a.sh, but replacing Norm-OPGRU with LSTMP.
# Copyright 2017 University of Chinese Academy of Sciences (UCAS) Gaofeng Cheng
# Apache 2.0

# Same as run_tdnn_opgru_1a.sh, but replacing Norm-OPGRU with LSTMP.
# Also Batchnorm in TDNN layers does not reduce the WER in Fisher+SWBD, so in run_tdnn_lstm_1a.sh,
# I just apply renorm component in TDNN layers.
# ./local/chain/compare_wer_general.sh --looped tdnn_lstm_1a_sp
# System tdnn_lstm_1a_sp
# WER on eval2000(tg) 12.3
Expand All @@ -15,15 +20,15 @@
# Final train prob (xent) -0.882
# Final valid prob (xent) -0.9393

# ./show_chain_wer.sh tdnn_lstm_1b_sp
# ./show_chain_wer.sh tdnn_lstm_1a_sp
# %WER 16.0 | 2628 21594 | 86.3 9.0 4.7 2.3 16.0 54.4 | exp/chain/tdnn_lstm_1a_sp/decode_eval2000_fsh_sw1_tg/score_7_0.0/eval2000_hires.ctm.callhm.filt.sys
# %WER 12.3 | 4459 42989 | 89.4 7.1 3.5 1.7 12.3 49.8 | exp/chain/tdnn_lstm_1a_sp/decode_eval2000_fsh_sw1_tg/score_8_0.0/eval2000_hires.ctm.filt.sys
# %WER 8.4 | 1831 21395 | 92.7 5.1 2.2 1.1 8.4 42.3 | exp/chain/tdnn_lstm_1a_sp/decode_eval2000_fsh_sw1_tg/score_10_0.0/eval2000_hires.ctm.swbd.filt.sys
# %WER 15.9 | 2628 21594 | 86.4 8.9 4.7 2.3 15.9 54.3 | exp/chain/tdnn_lstm_1a_sp/decode_eval2000_fsh_sw1_fg/score_7_0.0/eval2000_hires.ctm.callhm.filt.sys
# %WER 12.1 | 4459 42989 | 89.6 6.9 3.5 1.7 12.1 49.2 | exp/chain/tdnn_lstm_1a_sp/decode_eval2000_fsh_sw1_fg/score_8_0.0/eval2000_hires.ctm.filt.sys
# %WER 8.2 | 1831 21395 | 93.1 5.1 1.8 1.3 8.2 41.7 | exp/chain/tdnn_lstm_1a_sp/decode_eval2000_fsh_sw1_fg/score_8_0.0/eval2000_hires.ctm.swbd.filt.sys

# ./show_chain_wer_rt03.sh tdnn_lstm_1b_sp
# ./show_chain_wer_rt03.sh tdnn_lstm_1a_sp
# %WER 9.6 | 3970 36721 | 91.5 5.5 3.0 1.1 9.6 41.2 | exp/chain/tdnn_lstm_1a_sp/decode_rt03_fsh_sw1_tg/score_7_0.0/rt03_hires.ctm.fsh.filt.sys
# %WER 11.6 | 8420 76157 | 89.7 6.8 3.4 1.4 11.6 43.0 | exp/chain/tdnn_lstm_1a_sp/decode_rt03_fsh_sw1_tg/score_7_0.0/rt03_hires.ctm.filt.sys
# %WER 13.3 | 4450 39436 | 88.0 7.4 4.6 1.3 13.3 44.5 | exp/chain/tdnn_lstm_1a_sp/decode_rt03_fsh_sw1_tg/score_9_0.0/rt03_hires.ctm.swbd.filt.sys
Expand Down
2 changes: 2 additions & 0 deletions egs/fisher_swbd/s5/local/chain/run_tdnn_opgru_1a.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,6 @@
#!/bin/bash
# Copyright 2017 University of Chinese Academy of Sciences (UCAS) Gaofeng Cheng
# Apache 2.0

# This is based on TDNN_LSTM_1b (from egs/swbd/s5c), but using the NormOPGRU to replace the LSTMP,
# and adding chunk-{left,right}-context-initial=0
Expand Down