Some modifications to the SRE16 v2 recipe #1986

phanisankar-nidadavolu · 2017-10-30T22:24:12Z

Removed utils/data/get_utt2num_frames.sh and added --write-num-frames option to local/nnet3/prepare_feats_for_egs.sh
Made corresponding changes in run.sh

IMPORTANT

Results I got
Using Adapted PLDA, EER: Pooled 9.071%, Tagalog 12.98%, Cantonese 5.234%

Results mentioned in the original run.sh:

EER: Pooled 8.57%, Tagalog 12.29%, Cantonese 4.89%

They are not the same. Are the results in run.sh updated or is it something that I did wrong (in the modified recipe)?

…en extracting the features

david-ryan-snyder · 2017-10-30T22:48:26Z

egs/sre16/v1/local/nnet3/xvector/tuning/run_xvector_1a.sh

@@ -127,7 +127,7 @@ fi
 dropout_schedule='0,0@0.20,0.1@0.50,0'
 srand=123
 if [ $stage -le 6 ]; then
-  steps/nnet3/train_raw_dnn.py --stage=$train_stage \
+  python steps/nnet3/train_raw_dnn.py --stage=$train_stage \
    --cmd="$train_cmd" \
    --trainer.optimization.proportional-shrink 10 \
    --trainer.optimization.momentum=0.5 \


Is this change correct? I think the shebang at the top of the python file should take care of it.

It works but not necessary.

I think you should revert this change. It seems customary to rely on what's in the shebang.

For example, look at https://github.com/kaldi-asr/kaldi/blob/master/egs/swbd/s5c/local/chain/tuning/run_tdnn_7m.sh line 173

david-ryan-snyder · 2017-10-30T22:49:34Z

egs/sre16/v2/run.sh

@@ -178,10 +181,9 @@ if [ $stage -le 3 ]; then
  # wasteful, as it roughly doubles the amount of training data on disk.  After
  # creating training examples, this can be removed.
  local/nnet3/xvector/prepare_feats_for_egs.sh --nj 40 --cmd "$train_cmd" \
+    --write_utt2num_frames true \


Usually the options look like this: --write-utt2num-frames

My bad. Let me change it.

david-ryan-snyder · 2017-10-31T00:38:59Z

egs/sre16/v1/local/nnet3/xvector/prepare_feats_for_egs.sh

@@ -53,18 +65,31 @@ for n in $(seq $nj); do
  utils/create_data_link.pl $featdir/xvector_feats_${name}.$n.ark
 done


Looks like the call to create_data_link.pl is invoked at line 55 and again at 65. I think you should delete the first one.

Also, I don't think you need to do the create_split_dir.pl here. I think that should be the responsibility of the caller.

david-ryan-snyder

Thanks! There were just a few typos though. See inline comments.

david-ryan-snyder · 2017-10-31T13:59:55Z

egs/sre16/v1/local/nnet3/xvector/prepare_feats_for_egs.sh


 for n in $(seq $nj); do
-  # the next command does nothing unless $featdir/storage/ exists, see
+  # the next command does nothing unless $mfccdir/storage/ exists, see


This should say featdir, not mfccdir

david-ryan-snyder · 2017-10-31T14:00:52Z

egs/sre16/v1/local/nnet3/xvector/tuning/run_xvector_1a.sh

@@ -113,7 +113,7 @@ if [ $stage -le 5 ]; then
  output-layer name=output include-log-softmax=true dim=${num_targets}
 EOF

-  steps/nnet3/xconfig_to_configs.py \
+  python steps/nnet3/xconfig_to_configs.py \


Please change this back to the way it was... there are reasons to rely on what's in the shebang line of the python file.

david-ryan-snyder · 2017-10-31T14:01:09Z

egs/sre16/v2/run.sh

@@ -28,7 +28,7 @@ sre16_trials_tgl=data/sre16_eval_test/trials_tgl
 sre16_trials_yue=data/sre16_eval_test/trials_yue
 nnet_dir=exp/xvector_nnet_1a

-stage=0
+stage=10


Should be stage=0

david-ryan-snyder · 2017-10-31T18:46:55Z

Thanks Phani! This looks good to me. @danpovey, I think we can merge it.

danpovey

A couple additional comments

danpovey · 2017-10-31T18:49:43Z

egs/sre16/v1/local/nnet3/xvector/prepare_feats_for_egs.sh

@@ -13,6 +13,7 @@ norm_vars=false
 center=true
 compress=true
 cmn_window=300
+write_utt2num_frames=false  # if true writes utt2num_frames


Is there any circumstance under which you don't need this option? It seems to me it would be better to not have an option at all, and just write this in any case.

danpovey · 2017-10-31T18:50:07Z

egs/sre16/v1/local/nnet3/xvector/prepare_feats_for_egs.sh

@@ -41,30 +42,48 @@ done
 # Set various variables.
 mkdir -p $dir/log
 mkdir -p $data_out
-featdir=${PWD}/$dir
+featdir=`readlink -f $dir`


This command is not portable to BSD. Use utils/make_absolute.sh instead of readlink of.
Also we prefer $() to backticks.

david-ryan-snyder · 2017-10-31T21:26:39Z

egs/sre16/v1/local/nnet3/xvector/prepare_feats_for_egs.sh

@@ -41,30 +41,42 @@ done
 # Set various variables.
 mkdir -p $dir/log
 mkdir -p $data_out
-featdir=${PWD}/$dir
+featdir=`utils/make_absolute.sh $dir`


Don't forget what Dan said:

Also we prefer $() to backticks.

david-ryan-snyder · 2017-10-31T22:00:43Z

Thanks. @danpovey, I'm pretty sure it's ready now.

@christophfeinauer

* 'master' of https://github.com/kaldi-asr/kaldi: [src,scripts] nnet1-related changes: (kaldi-asr#1998) [egs] rotation for image augmentation in CIFAR example (kaldi-asr#1955) [egs] Minor fixes to the SRE16 v2 recipe (kaldi-asr#1986) [egs] Remove deprecated non-working scripts [scripts] Fixes to segment_long_utterances.sh (thanks @christophfeinauer) and train_raw_dnn.py (kaldi-asr#1993) [src] Minor fix: change to error message (kaldi-asr#1980) [egs] Add example of component-level l2-regularize for WSJ scripts [egs] Small fix to Chime4 RE data location (kaldi-asr#1966) [build] Remove download of ATLAS header files from tools/. (kaldi-asr#1974)

Added write-utt2num-frames option for getting the number of frames wh…

b19c1a2

…en extracting the features

david-ryan-snyder reviewed Oct 30, 2017

View reviewed changes

david-ryan-snyder reviewed Oct 31, 2017

View reviewed changes

Made some modifications that David suggested

0db3716

david-ryan-snyder reviewed Oct 31, 2017

View reviewed changes

Fixed the stage and other changes that David suggested

6d0cb4c

danpovey reviewed Oct 31, 2017

View reviewed changes

Made changes based on Dan's comments

2cebce7

david-ryan-snyder reviewed Oct 31, 2017

View reviewed changes

Removed backtick

aff609a

danpovey merged commit 3ea5340 into kaldi-asr:master Nov 2, 2017

xiaohui-zhang pushed a commit to xiaohui-zhang/kaldi that referenced this pull request Nov 14, 2017

[egs] Minor fixes to the SRE16 v2 recipe (kaldi-asr#1986)

c89912a

mahsa7823 pushed a commit to mahsa7823/kaldi that referenced this pull request Feb 28, 2018

[egs] Minor fixes to the SRE16 v2 recipe (kaldi-asr#1986)

d6821e1

Skaiste pushed a commit to Skaiste/idlak that referenced this pull request Sep 26, 2018

[egs] Minor fixes to the SRE16 v2 recipe (kaldi-asr#1986)

a64aefd

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Some modifications to the SRE16 v2 recipe #1986

Some modifications to the SRE16 v2 recipe #1986

phanisankar-nidadavolu commented Oct 30, 2017

david-ryan-snyder Oct 30, 2017

phanisankar-nidadavolu Oct 30, 2017

david-ryan-snyder Oct 30, 2017

phanisankar-nidadavolu Oct 30, 2017

david-ryan-snyder Oct 30, 2017

david-ryan-snyder Oct 30, 2017

phanisankar-nidadavolu Oct 30, 2017

david-ryan-snyder Oct 31, 2017

david-ryan-snyder left a comment

david-ryan-snyder Oct 31, 2017

david-ryan-snyder Oct 31, 2017

david-ryan-snyder Oct 31, 2017

david-ryan-snyder commented Oct 31, 2017

danpovey left a comment

danpovey Oct 31, 2017

danpovey Oct 31, 2017

david-ryan-snyder Oct 31, 2017

david-ryan-snyder commented Oct 31, 2017

		@@ -53,18 +65,31 @@ for n in $(seq $nj); do
		utils/create_data_link.pl $featdir/xvector_feats_${name}.$n.ark
		done

Some modifications to the SRE16 v2 recipe #1986

Some modifications to the SRE16 v2 recipe #1986

Conversation

phanisankar-nidadavolu commented Oct 30, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

david-ryan-snyder left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

david-ryan-snyder commented Oct 31, 2017

danpovey left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

david-ryan-snyder commented Oct 31, 2017