Running mini-librispeech kws locally

[Ryan Woo]

Hello,

I am a Kaldi beginner. I'm trying to run the mini-librispeech kws scripts on my computer, and I'm hitting a snag with some paths that do not exist. I started by editing cmd.sh to use run.pl, and then running run.sh. That completed successfully, and now I'm trying to run run_kws.sh.

(The === BEGIN STAGE X === are just echo statements I added to help me see where failures are occurring)

(.venv) rwoo@ryan-clam:~/Documents/columbia/coms6998-FundamentalsOfSpeechRecognition/kaldi/egs/mini_librispeech/s5$ ./local/kws/run_kws.sh
==== ./local/kws/run_kws.sh BEGIN STAGE 1 ====
Number of trials: 7328.96
Frame lengths: 0.01
Generating map files
Found 0 phrases containing (at least one) OOV...
==== ./local/kws/run_kws.sh END STAGE 1 ====
==== ./local/kws/run_kws.sh BEGIN STAGE 2 ====
steps/align_fmllr.sh --nj 5 --cmd run.pl data/dev_clean_2 data/lang exp/tri3b exp/tri3b_ali_dev_clean_2
steps/align_fmllr.sh: feature type is lda
steps/align_fmllr.sh: compiling training graphs
steps/align_fmllr.sh: aligning data in data/dev_clean_2 using exp/tri3b/final.alimdl and speaker-independent features.
steps/align_fmllr.sh: computing fMLLR transforms
steps/align_fmllr.sh: doing final alignment.
steps/align_fmllr.sh: done aligning data.
steps/diagnostic/analyze_alignments.sh --cmd run.pl data/lang exp/tri3b_ali_dev_clean_2
steps/diagnostic/analyze_alignments.sh: see stats in exp/tri3b_ali_dev_clean_2/log/analyze_alignments.log
7 warnings in exp/tri3b_ali_dev_clean_2/log/align_pass1.*.log
5 warnings in exp/tri3b_ali_dev_clean_2/log/align_pass2.*.log
local/kws/create_hitlist.sh: writing alignments.
local/kws/create_hitlist.sh: done generating hitlist
==== ./local/kws/run_kws.sh END STAGE 2 ====
==== ./local/kws/run_kws.sh BEGIN STAGE 3 ====
local/kws/compile_keywords.sh data/dev_clean_2/kws/ data/lang data/dev_clean_2/kws//tmp.2
transcripts-to-fsts ark:data/dev_clean_2/kws//tmp.2/keywords.int ark,scp,t:data/dev_clean_2/kws//tmp.2/keywords.fsts,-
LOG (transcripts-to-fsts[5.5.1001~1-7460d]:main():transcripts-to-fsts.cc:159) Done 6 transcriptions
==== ./local/kws/run_kws.sh END STAGE 3 ====
==== ./local/kws/run_kws.sh BEGIN STAGE 4 ====
steps/make_index.sh --cmd run.pl --lmwt 8 --acwt 1.0 --frame-subsampling-factor 3 data/dev_clean_2/kws/ data/lang exp/chain/tdnn1h_sp_online/decode_tglarge_dev_clean_2/ exp/chain/tdnn1h_sp_online/decode_tglarge_dev_clean_2//kws_indices_8
cat: exp/chain/tdnn1h_sp_online/decode_tglarge_dev_clean_2//num_jobs: No such file or directory

Are there additional configurations I need to set in order for these files to appear? I tried to just fill in the num_jobs file, but it fails closely after with 

steps/make_index.sh: Error: no such file exp/chain/tdnn1h_sp_online/final.mdl

I do have something in the chain2 directory if it is related

(.venv) rwoo@ryan-clam:~/Documents/columbia/coms6998-FundamentalsOfSpeechRecognition/kaldi/egs/mini_librispeech/s5$ ls exp/chain2/tdnn1a_sp/
0_trans.mdl  5.raw  cache.6  decode_tglarge_dev_clean_2  default.tree  egs                final.mdl  final_train.mdl  lda.mat  processed_egs  schedule.txt
4.raw        6.raw  configs  decode_tgsmall_dev_clean_2  den_fsts      final_heldout.mdl  final.raw  init             log      raw_egs
(.venv) rwoo@ryan-clam:~/Documents/columbia/coms6998-FundamentalsOfSpeechRecognition/kaldi/egs/mini_librispeech/s5$ 

Thanks

[aliiire...@gmail.com]

hi

You must change the model directory in this line:

https://github.com/kaldi-asr/kaldi/blob/e4940d045d39deb86016bc176893303b5240ff59/egs/mini_librispeech/s5/local/kws/run_kws.sh#L86