`ali_dir` defined in `run_blstm_6h.sh` (fisher_swbd) not available

I'm in the process of replicating the BLSTM model in fisher_swbd. I have completed run.sh and local/nnet3/run_ivector_common.sh. Upon running local/chain/run_blstm_6h.sh, I find that the alignment directory defined here isn't generated by any other scripts. Am I missing something?

That would be generated by the top-level run.sh

On Mon, Jan 2, 2017 at 12:10 PM, Z. Stanley Guan notifications@github.com wrote:

I'm in the process of replicating the BLSTM model in fisher_swbd. I have completed run.sh and local/nnet3/run_ivector_common.sh. Upon running local/chain/run_blstm_6h.sh, I find that the alignment directory defined here https://github.com/kaldi-asr/kaldi/blob/master/egs/fisher_swbd/s5/local/chain/run_blstm_6h.sh#L41 isn't generated by any other scripts. Am I missing something?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVu6MOzR8EwEOEmyk_0hnzD4KUerNEks5rOVmkgaJpZM4LZMOS .

I see tri_5a_ali but not tri_5a_ali_nodup in the top level run.sh

Hm, yes, I see. The easy fix is to just to remove the _nodup; it seems that we just named the alignment directory differently in the top-level run.sh @naxingyu, you created these scripts; can you please look into this and fix it appropriately?

I did try to use tri5a_ali instead of tri5a_ali_nodup as ali_dir (duplicating the former and naming that the latter). But for the tree-building step I got the following error. My kaldi is at commit ba58e8e55ed1777969e670980dbbc17c0eef530c.

Is tri5a_ali the right directory to use, or do I actually need to generate alignments for the speed-perturbed data?

Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 --leftmost-questions-truncate -1 --cmd run.pl 11000 data/train_nodup_sp data/lang_chain exp/tri5a_ali_nodup exp/chain/tri6_tree_11000
Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: feature type is lda
Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Using transforms from exp/tri5a_ali_nodup
Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Initializing monophone model (for alignment conversion, in case topology changed)
feat-to-dim 'ark,s,cs:apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark$
transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:-
transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:-
apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:-
splice-feats --left-context=3 --right-context=3 ark:- ark:-
subsample-feats --n=3 ark:- ark:-
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000376-000554, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000582-000648, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001005-001088, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001182-001231, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001447-001573, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-002019-002125, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003133-003258, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003714-004049, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004072-004162, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004221-004403, producing no output for this utterance

######## (similar messages in between omitted) ########

WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-045799-045988, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-047217-047411, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051469-051741, producing no output for this utterance
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051770-051944, producing no output for this utterance
LOG (apply-cmvn:main():apply-cmvn.cc:146) Applied cepstral mean normalization to 52509 utterances, errors on 0
WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-052010-052276, producing no output for this utterance
LOG (transform-feats:main():transform-feats.cc:158) Overall average [pseudo-]logdet is -90.0966 over 1.79498e+07 frames.
LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 52509 utterances; 0 had errors.
LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 0 utterances; 52509 had errors.
LOG (subsample-feats:main():subsample-feats.cc:115) Processed 0 feature matrices; 0 with errors.
LOG (subsample-feats:main():subsample-feats.cc:117) Processed 0 input frames and 0 output frames.
ERROR (feat-to-dim:main():feat-to-dim.cc:58) Could not read any features (empty archive?)

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*)
kaldi::MessageLogger::~MessageLogger()
main
__libc_start_main
feat-to-dim() [0x4127b9]

WARNING (feat-to-dim:Close():kaldi-io.cc:501) Pipe apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- | had nonzero return status 256
ERROR (feat-to-dim:~SequentialTableReaderArchiveImpl():util/kaldi-table-inl.h:678) TableReader: error detected closing archive 'apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- |'

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*)
kaldi::MessageLogger::~MessageLogger()
kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix<float> > >::~SequentialTableReaderArchiveImpl()
kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix<float> > >::~SequentialTableReaderArchiveImpl()
kaldi::SequentialTableReader<kaldi::KaldiObjectHolder<kaldi::Matrix<float> > >::~SequentialTableReader()
main
__libc_start_main
feat-to-dim() [0x4127b9]

steps/nnet3/chain/build_tree.sh: line 116: 62579 Aborted                 (core dumped) feat-to-dim "$example_feats" -
Jan 02 12:07:11 error getting feature dimension

I didn't make these scripts, but I do see that your data directory is "data/train_nodup_sp" while the alignment directory is "exp/tri5a_ali_nodup". Based on the error message, I'm pretty sure your suspicion is correct: you need to either make alignments for the speed perturbed data (i.e., make tri5a_ali_nodup_sp) with the model in tri5a, or you need to use "data/train_nodup" (instead of "data/train_nodup_sp") with "tri5_ali_nodup". Probably, the intent of the script is to do the former, as speed perturbation tends to always help training generalization.

On Mon, Jan 2, 2017 at 4:19 PM, Z. Stanley Guan notifications@github.com wrote:

I did try to use tri5a_ali instead of tri5a_ali_nodup as ali_dir (duplicating the former and naming that the latter). But for the tree-building step https://github.com/kaldi-asr/kaldi/blob/master/egs/fisher_swbd/s5/local/chain/run_blstm_6h.sh#L76 I got the following error. My kaldi is at commit ba58e8e https://github.com/kaldi-asr/kaldi/commit/ba58e8e55ed1777969e670980dbbc17c0eef530c .

Is tri5a_ali the right directory to use, or do I actually need to generate alignments for the speed-perturbed data?

Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 --leftmost-questions-truncate -1 --cmd run.pl 11000 data/train_nodup_sp data/lang_chain exp/tri5a_ali_nodup exp/chain/tri6_tree_11000 Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: feature type is lda Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Using transforms from exp/tri5a_ali_nodup Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Initializing monophone model (for alignment conversion, in case topology changed) feat-to-dim 'ark,s,cs:apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark$ transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- splice-feats --left-context=3 --right-context=3 ark:- ark:- subsample-feats --n=3 ark:- ark:- WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000376-000554, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000582-000648, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001005-001088, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001182-001231, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001447-001573, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-002019-002125, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003133-003258, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003714-004049, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004072-004162, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004221-004403, producing no output for this utterance

######## (similar messages in between omitted) ########

WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-045799-045988, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-047217-047411, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051469-051741, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051770-051944, producing no output for this utterance LOG (apply-cmvn:main():apply-cmvn.cc:146) Applied cepstral mean normalization to 52509 utterances, errors on 0 WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-052010-052276, producing no output for this utterance LOG (transform-feats:main():transform-feats.cc:158) Overall average [pseudo-]logdet is -90.0966 over 1.79498e+07 frames. LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 52509 utterances; 0 had errors. LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 0 utterances; 52509 had errors. LOG (subsample-feats:main():subsample-feats.cc:115) Processed 0 feature matrices; 0 with errors. LOG (subsample-feats:main():subsample-feats.cc:117) Processed 0 input frames and 0 output frames. ERROR (feat-to-dim:main():feat-to-dim.cc:58) Could not read any features (empty archive?)

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*) kaldi::MessageLogger::~MessageLogger() main __libc_start_main feat-to-dim() [0x4127b9]

WARNING (feat-to-dim:Close():kaldi-io.cc:501) Pipe apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- | had nonzero return status 256 ERROR (feat-to-dim:~SequentialTableReaderArchiveImpl():util/kaldi-table-inl.h:678) TableReader: error detected closing archive 'apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- |'

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*) kaldi::MessageLogger::~MessageLogger() kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix > >::~SequentialTableReaderArchiveImpl() kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix > >::~SequentialTableReaderArchiveImpl() kaldi::SequentialTableReader<kaldi::KaldiObjectHolder<kaldi::Matrix > >::~SequentialTableReader() main __libc_start_main feat-to-dim() [0x4127b9]

steps/nnet3/chain/build_tree.sh: line 116: 62579 Aborted (core dumped) feat-to-dim "$example_feats" - Jan 02 12:07:11 error getting feature dimension

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270035734, or mute the thread https://github.com/notifications/unsubscribe-auth/AEi_UDxYVgVw0QvgkAYvMElU-cu5fEYzks5rOZP9gaJpZM4LZMOS .

-- Daniel Galvez

An easier fix might be to link tri5a_ali_nodup to tri5a_ali, which does exist. Xingyu will fix this in the proper way when he has a chance.

On Mon, Jan 2, 2017 at 5:00 PM, Daniel Galvez notifications@github.com wrote:

I didn't make these scripts, but I do see that your data directory is "data/train_nodup_sp" while the alignment directory is "exp/tri5a_ali_nodup". Based on the error message, I'm pretty sure your suspicion is correct: you need to either make alignments for the speed perturbed data (i.e., make tri5a_ali_nodup_sp) with the model in tri5a, or you need to use "data/train_nodup" (instead of "data/train_nodup_sp") with "tri5_ali_nodup". Probably, the intent of the script is to do the former, as speed perturbation tends to always help training generalization.

On Mon, Jan 2, 2017 at 4:19 PM, Z. Stanley Guan notifications@github.com wrote:

I did try to use tri5a_ali instead of tri5a_ali_nodup as alidir (duplicating the former and naming that the latter). But for the tree-building step <https://github.com/kaldi-asr/kaldi/blob/master/egs/fisher swbd/s5/local/chain/run_blstm_6h.sh#L76> I got the following error. My kaldi is at commit ba58e8e https://github.com/kaldi-asr/kaldi/commit/ ba58e8e55ed1777969e670980dbbc17c0eef530c

.

Is tri5a_ali the right directory to use, or do I actually need to generate alignments for the speed-perturbed data?

Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 --leftmost-questions-truncate -1 --cmd run.pl 11000 data/train_nodup_sp data/lang_chain exp/tri5a_ali_nodup exp/chain/tri6_tree_11000 Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: feature type is lda Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Using transforms from exp/tri5a_ali_nodup Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Initializing monophone model (for alignment conversion, in case topology changed) feat-to-dim 'ark,s,cs:apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark$ transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- splice-feats --left-context=3 --right-context=3 ark:- ark:- subsample-feats --n=3 ark:- ark:- WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000376-000554, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000582-000648, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001005-001088, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001182-001231, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001447-001573, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-002019-002125, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003133-003258, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003714-004049, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004072-004162, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004221-004403, producing no output for this utterance

######## (similar messages in between omitted) ########

WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-045799-045988, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-047217-047411, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051469-051741, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051770-051944, producing no output for this utterance LOG (apply-cmvn:main():apply-cmvn.cc:146) Applied cepstral mean normalization to 52509 utterances, errors on 0 WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-052010-052276, producing no output for this utterance LOG (transform-feats:main():transform-feats.cc:158) Overall average [pseudo-]logdet is -90.0966 over 1.79498e+07 frames. LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 52509 utterances; 0 had errors. LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 0 utterances; 52509 had errors. LOG (subsample-feats:main():subsample-feats.cc:115) Processed 0 feature matrices; 0 with errors. LOG (subsample-feats:main():subsample-feats.cc:117) Processed 0 input frames and 0 output frames. ERROR (feat-to-dim:main():feat-to-dim.cc:58) Could not read any features (empty archive?)

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*) kaldi::MessageLogger::~MessageLogger() main __libc_start_main feat-to-dim() [0x4127b9]

WARNING (feat-to-dim:Close():kaldi-io.cc:501) Pipe apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- | had nonzero return status 256 ERROR (feat-to-dim:~SequentialTableReaderArchiveIm pl():util/kaldi-table-inl.h:678) TableReader: error detected closing archive 'apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- |'

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*) kaldi::MessageLogger::~MessageLogger() kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix

::~SequentialTableReaderArchiveImpl() kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix ::~SequentialTableReaderArchiveImpl() kaldi::SequentialTableReader<kaldi::KaldiObjectHolder<kaldi::Matrix ::~SequentialTableReader() main __libc_start_main feat-to-dim() [0x4127b9]

steps/nnet3/chain/build_tree.sh: line 116: 62579 Aborted (core dumped) feat-to-dim "$example_feats" - Jan 02 12:07:11 error getting feature dimension

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270035734, or mute the thread https://github.com/notifications/unsubscribe- auth/AEi_UDxYVgVw0QvgkAYvMElU-cu5fEYzks5rOZP9gaJpZM4LZMOS .

-- Daniel Galvez

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270038136, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVu-FBBNlFknt-yLVyySP3FjQTIMz5ks5rOZ3KgaJpZM4LZMOS .

On it.

在 2017年1月3日，10:05，Daniel Povey notifications@github.com 写道：

An easier fix might be to link tri5a_ali_nodup to tri5a_ali, which does exist. Xingyu will fix this in the proper way when he has a chance.

On Mon, Jan 2, 2017 at 5:00 PM, Daniel Galvez notifications@github.com wrote:

I didn't make these scripts, but I do see that your data directory is "data/train_nodup_sp" while the alignment directory is "exp/tri5a_ali_nodup". Based on the error message, I'm pretty sure your suspicion is correct: you need to either make alignments for the speed perturbed data (i.e., make tri5a_ali_nodup_sp) with the model in tri5a, or you need to use "data/train_nodup" (instead of "data/train_nodup_sp") with "tri5_ali_nodup". Probably, the intent of the script is to do the former, as speed perturbation tends to always help training generalization.

On Mon, Jan 2, 2017 at 4:19 PM, Z. Stanley Guan notifications@github.com wrote:

I did try to use tri5a_ali instead of tri5a_ali_nodup as alidir (duplicating the former and naming that the latter). But for the tree-building step <https://github.com/kaldi-asr/kaldi/blob/master/egs/fisher swbd/s5/local/chain/run_blstm_6h.sh#L76> I got the following error. My kaldi is at commit ba58e8e https://github.com/kaldi-asr/kaldi/commit/ ba58e8e55ed1777969e670980dbbc17c0eef530c

.

Is tri5a_ali the right directory to use, or do I actually need to generate alignments for the speed-perturbed data?

Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 --leftmost-questions-truncate -1 --cmd run.pl 11000 data/train_nodup_sp data/lang_chain exp/tri5a_ali_nodup exp/chain/tri6_tree_11000 Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: feature type is lda Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Using transforms from exp/tri5a_ali_nodup Jan 02 12:06:00 steps/nnet3/chain/build_tree.sh: Initializing monophone model (for alignment conversion, in case topology changed) feat-to-dim 'ark,s,cs:apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark$ transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- splice-feats --left-context=3 --right-context=3 ark:- ark:- subsample-feats --n=3 ark:- ark:- WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000376-000554, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-000582-000648, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001005-001088, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001182-001231, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-001447-001573, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-002019-002125, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003133-003258, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-003714-004049, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004072-004162, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00001-A-004221-004403, producing no output for this utterance

######## (similar messages in between omitted) ########

WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-045799-045988, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-047217-047411, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051469-051741, producing no output for this utterance WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-051770-051944, producing no output for this utterance LOG (apply-cmvn:main():apply-cmvn.cc:146) Applied cepstral mean normalization to 52509 utterances, errors on 0 WARNING (transform-feats:main():transform-feats.cc:87) No fMLLR transform available for utterance sp0.9-fe_03_00330-A-052010-052276, producing no output for this utterance LOG (transform-feats:main():transform-feats.cc:158) Overall average [pseudo-]logdet is -90.0966 over 1.79498e+07 frames. LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 52509 utterances; 0 had errors. LOG (transform-feats:main():transform-feats.cc:161) Applied transform to 0 utterances; 52509 had errors. LOG (subsample-feats:main():subsample-feats.cc:115) Processed 0 feature matrices; 0 with errors. LOG (subsample-feats:main():subsample-feats.cc:117) Processed 0 input frames and 0 output frames. ERROR (feat-to-dim:main():feat-to-dim.cc:58) Could not read any features (empty archive?)

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*) kaldi::MessageLogger::~MessageLogger() main __libc_start_main feat-to-dim() [0x4127b9]

WARNING (feat-to-dim:Close():kaldi-io.cc:501) Pipe apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- | had nonzero return status 256 ERROR (feat-to-dim:~SequentialTableReaderArchiveIm pl():util/kaldi-table-inl.h:678) TableReader: error detected closing archive 'apply-cmvn --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk scp:data/train_nodup_sp/split100/1/cmvn.scp scp:data/train_nodup_sp/split100/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri5a_ali_nodup/final.mat ark:- ark:- | transform-feats --utt2spk=ark:data/train_nodup_sp/split100/1/utt2spk ark,s,cs:exp/tri5a_ali_nodup/trans.1 ark:- ark:- | subsample-feats --n=3 ark:- ark:- |'

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*) kaldi::MessageLogger::~MessageLogger() kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix

::~SequentialTableReaderArchiveImpl() kaldi::SequentialTableReaderArchiveImpl<kaldi::KaldiObjectHolder<kaldi::Matrix ::~SequentialTableReaderArchiveImpl() kaldi::SequentialTableReader<kaldi::KaldiObjectHolder<kaldi::Matrix ::~SequentialTableReader() main __libc_start_main feat-to-dim() [0x4127b9]

steps/nnet3/chain/build_tree.sh: line 116: 62579 Aborted (core dumped) feat-to-dim "$example_feats" - Jan 02 12:07:11 error getting feature dimension

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270035734, or mute the thread https://github.com/notifications/unsubscribe- auth/AEi_UDxYVgVw0QvgkAYvMElU-cu5fEYzks5rOZP9gaJpZM4LZMOS .

-- Daniel Galvez

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270038136, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVu-FBBNlFknt-yLVyySP3FjQTIMz5ks5rOZ3KgaJpZM4LZMOS .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270041901, or mute the thread https://github.com/notifications/unsubscribe-auth/ADKpxD3cY5_9RoMRtPW5x9lZO5bKpvS0ks5rOazdgaJpZM4LZMOS.

@danpovey I'll remove '_nodup' from the experiments directories, and fix some dir naming problems that I didn't notice. I believe there won't be a compatibility issue.

Is the fix simply removing _nodup? I got the error above when using tri5a_ali.

I think you must have done something else that was wrong when removing _nodup, because your command seems to be mixing things that have the _sp suffix (which uses speed-perturbed data), with things that don't; and that isn't right.

On Mon, Jan 2, 2017 at 7:21 PM, Z. Stanley Guan notifications@github.com wrote:

Is the fix simply removing _nodup? I got the error above when using tri5a_ali.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270046578, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVuwkQuXEWviS20HETFNV0N8vc8ixyks5rOb7MgaJpZM4LZMOS .

The current script is using speed-perturbed data and tri5a_ali isn't generated with that.

Oh yes, you're right. Xingyu will figure out what went wrong and fix it.

On Mon, Jan 2, 2017 at 7:29 PM, Z. Stanley Guan notifications@github.com wrote:

The current script is using speed-perturbed data and tri5a_ali isn't generated with that.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270047080, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVu162udlIuelWZshHm4S0WQPVhlfvks5rOcCAgaJpZM4LZMOS .

@stanleyguan hope #1303 resolves your problem.

So it looks like we do need to generate alignments here?

Yes I think that's possible. Xingyu, I suspect that in the script you copied this from, those alignments were not needed because the tree was being built on the original non-perturbed data (i.e. not the _sp data). But this script seem to be building the tree on the full perturbed data, which would require those alignments to be dumped.

On Mon, Jan 2, 2017 at 8:29 PM, Z. Stanley Guan notifications@github.com wrote:

So it looks like we do need to generate alignments here https://github.com/kaldi-asr/kaldi/blob/master/egs/fisher_swbd/s5/local/chain/run_blstm_6h.sh#L50 ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270050741, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVu8zPw-rjwOupbgzEqfn9L7zB8Qo4ks5rOc7AgaJpZM4LZMOS .

Yes, I made a mistake in the previous PR. I’ve committed a new one to fix it. Now I remember everything :)

在 2017年1月3日，12:44，Daniel Povey notifications@github.com 写道：

Yes I think that's possible. Xingyu, I suspect that in the script you copied this from, those alignments were not needed because the tree was being built on the original non-perturbed data (i.e. not the _sp data). But this script seem to be building the tree on the full perturbed data, which would require those alignments to be dumped.

On Mon, Jan 2, 2017 at 8:29 PM, Z. Stanley Guan notifications@github.com wrote:

So it looks like we do need to generate alignments here https://github.com/kaldi-asr/kaldi/blob/master/egs/fisher_swbd/s5/local/chain/run_blstm_6h.sh#L50 ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270050741, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVu8zPw-rjwOupbgzEqfn9L7zB8Qo4ks5rOc7AgaJpZM4LZMOS .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kaldi-asr/kaldi/issues/1299#issuecomment-270051555, or mute the thread https://github.com/notifications/unsubscribe-auth/ADKpxIRuxw82yyqQ7UXj6cfN1moBgVUXks5rOdIpgaJpZM4LZMOS.

I think it's now resolved.

kaldi-asr / kaldi

`ali_dir` defined in `run_blstm_6h.sh` (fisher_swbd) not available #1299