ContextFst: CreateArc, invalid olabel supplied [confusion about phone list or disambig symbols?] error in using LSTM

411 views
Skip to first unread message

Masoud Abasian

unread,
Jun 26, 2016, 6:34:01 AM6/26/16
to kaldi-help
Hello.
i want to use LSTM for "farsdat" database with nnet3 scripts in "swbd" database.

I have an error in this way

in local/nnet3/run_lstm.sh :
local/nnet3/run_ivector_common.sh --stage $stage \
  --speed-perturb $speed_perturb || exit 1;

in local/nnet3/run_ivector_common.sh :
 steps/align_fmllr.sh --nj 49 --cmd "$train_cmd" \
      data/train data/lang exp/sgmm2_4 exp/tri3_ali_nodup_sp || exit 1


in
steps/align_fmllr.sh   
$cmd JOB=1:$nj $dir/log/compile_graphs.JOB.log  \
      compile-train-graphs $dir/tree $dir/final.mdl  $lang/L.fst "$tra" \
        "ark:|gzip -c >$dir/fsts.JOB.gz" || exit 1;


and the log error said :

oov.txt file or word.txt have any problem while before LSTM, they don't have problem? or
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# compile-train-graphs exp/tri3_ali_nodup_sp/tree exp/tri3_ali_nodup_sp/final.mdl data/lang/L.fst "ark:utils/sym2int.pl --map-oov 28 -f 2- data/lang/words.txt data/train/split49/1/text|" "ark:|gzip -c >exp/tri3_ali_nodup_sp/fsts.1.gz"
# Started at Sun Jun 26 14:52:40 IRDT 2016
#
compile-train-graphs exp/tri3_ali_nodup_sp/tree exp/tri3_ali_nodup_sp/final.mdl data/lang/L.fst 'ark:utils/sym2int.pl --map-oov 28 -f 2- data/lang/words.txt data/train/split49/1/text|' 'ark:|gzip -c >exp/tri3_ali_nodup_sp/fsts.1.gz'
ContextFst: CreateArc, invalid olabel supplied [confusion about phone list or disambig symbols?]: 109# Accounting: time=0 threads=1
# Ended (code 1) at Sun Jun 26 14:52:40 IRDT 2016, elapsed time 0 seconds
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

Why this error occur ? Is there any problem in oov and word.txt or in transcript text file in data/train/splite49 directory?

Daniel Povey

unread,
Jun 26, 2016, 2:28:44 PM6/26/16
to kaldi-help
I suspect you changed the phone list (e.g. adding position dependent
phones) and you are still somehow accessing models that use the old
phone list. Check your script.
Dan
> --
> You received this message because you are subscribed to the Google Groups
> "kaldi-help" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to kaldi-help+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

Masoud Abasian

unread,
Jul 4, 2016, 5:19:20 PM7/4/16
to kaldi-help, dpo...@gmail.com
Hi. I'm not sure, but i think this problem causes for difference version of kaldi in my system (which my code are written in this system and use new version of codes and scripts) and other system (which i run my codes on it and have GPU) . I didn't know that another system have old version of kaldi.
In my system arpa2fst.cc scrip have 146 line with so many option that I use this option , but on the other system arpa2fst.cc script have 62 lines and don't have some option (it's old script) .
So I update kaldi and this problem was solved.

Thank you for your help and responses.
Reply all
Reply to author
Forward
0 new messages