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

Jump to bottom

Replace LDA layer with delta and delta-delta features #3490

Merged
merged 8 commits into from
Aug 2, 2019
Merged

Replace LDA layer with delta and delta-delta features #3490

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
bb23ee9
added delta features script
desh2608 Jul 21, 2019
a634c34
added delta features script for swbd
desh2608 Jul 24, 2019
cc2d814
added delta feats chain script for wsj
desh2608 Jul 25, 2019
3947c1d
added results in the scripts
desh2608 Jul 25, 2019
b8e2509
added delta layer in xconfig
desh2608 Jul 26, 2019
ecb159d
small changes to WSJ and SWBD scripts
desh2608 Jul 27, 2019
056982b
Merge branch 'master' of https://github.com/kaldi-asr/kaldi into delt…
desh2608 Jul 28, 2019
9241679
made delta-layer compact by removing no-op-components; added mini_lib…
desh2608 Jul 28, 2019
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
Next Next commit
added delta features script
  • Loading branch information
@desh2608
desh2608 committed Jul 21, 2019
commit bb23ee927ca7cfea3e90cdc1b7e55a3446128356
16 changes: 13 additions & 3 deletions egs/swbd/s5c/local/chain/tuning/run_tdnn_7q.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ stage=0
train_stage=-10
get_egs_stage=-10
speed_perturb=true
affix=7q
affix=7r
if [ -e data/rt03 ]; then maybe_rt03=rt03; else maybe_rt03= ; fi

decode_iter=
Expand Down Expand Up @@ -131,13 +131,23 @@ if [ $stage -le 12 ]; then
input dim=100 name=ivector
input dim=40 name=input

no-op-component name=delta_1 input=Scale(-1.0, Offset(input, -1))
no-op-component name=delta_2 input=Scale(1.0, Offset(input, 1))
no-op-component name=delta_delta_1 input=Scale(-2.0, input)
no-op-component name=delta_delta_2 input=Scale(1.0, Offset(input,-2))
no-op-component name=delta_delta_3 input=Scale(1.0, Offset(input,2))

# please note that it is important to have input layer with the name=input
# as the layer immediately preceding the fixed-affine-layer to enable
# the use of short notation for the descriptor
fixed-affine-layer name=lda input=Append(-1,0,1,ReplaceIndex(ivector, t, 0)) affine-transform-file=$dir/configs/lda.mat
#fixed-affine-layer name=lda input=Append(-1,0,1,ReplaceIndex(ivector, t, 0)) affine-transform-file=$dir/configs/lda.mat
no-op-component name=input2 input=Append(Offset(input,0), Sum(delta_1, delta_2), Sum(delta_delta_1, delta_delta_2, delta_delta_3))

batchnorm-component name=bn_input
no-op-component name=input3 input=Append(bn_input, Scale(1.0, ReplaceIndex(ivector, t, 0)))

# the first splicing is moved before the lda layer, so no splicing here
relu-batchnorm-dropout-layer name=tdnn1 $affine_opts dim=1536
relu-batchnorm-dropout-layer name=tdnn1 $affine_opts dim=1536 input=input3
tdnnf-layer name=tdnnf2 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=1
tdnnf-layer name=tdnnf3 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=1
tdnnf-layer name=tdnnf4 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=1
Expand Down
16 changes: 13 additions & 3 deletions egs/wsj/s5/local/chain/tuning/run_tdnn_1g.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,7 @@ chunk_right_context=0

# training options
srand=0
remove_egs=true
remove_egs=false

#decode options
test_online_decoding=false # if true, it will run the last decoding stage.
Expand Down Expand Up @@ -182,13 +182,23 @@ if [ $stage -le 15 ]; then
input dim=100 name=ivector
input dim=40 name=input

no-op-component name=delta_1 input=Scale(-1.0, Offset(input, -1))
no-op-component name=delta_2 input=Scale(1.0, Offset(input, 1))
no-op-component name=delta_delta_1 input=Scale(-2.0, input)
no-op-component name=delta_delta_2 input=Scale(1.0, Offset(input,-2))
no-op-component name=delta_delta_3 input=Scale(1.0, Offset(input,2))

# please note that it is important to have input layer with the name=input
# as the layer immediately preceding the fixed-affine-layer to enable
# the use of short notation for the descriptor
fixed-affine-layer name=lda input=Append(-1,0,1,ReplaceIndex(ivector, t, 0)) affine-transform-file=$dir/configs/lda.mat
#fixed-affine-layer name=lda input=Append(-1,0,1,ReplaceIndex(ivector, t, 0)) affine-transform-file=$dir/configs/lda.mat
no-op-component name=input2 input=Append(Offset(input,0), Sum(delta_1, delta_2))

batchnorm-component name=bn_input
no-op-component name=input3 input=Append(bn_input, Scale(1.0, ReplaceIndex(ivector, t, 0)))

# the first splicing is moved before the lda layer, so no splicing here
relu-batchnorm-dropout-layer name=tdnn1 $tdnn_opts dim=1024
relu-batchnorm-dropout-layer name=tdnn1 $tdnn_opts dim=1024 input=input3
tdnnf-layer name=tdnnf2 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=1
tdnnf-layer name=tdnnf3 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=1
tdnnf-layer name=tdnnf4 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=1
Expand Down