bad variance error

[manjunath ke]

Hi all

I am trying to build monophone hmm models using kaldi. I get the following error. (Note: this error is clearly not due to lack of memory because I have considered single job)

manju@dinesh-pc:~/manju/asr/kaldi/my_exp/oracle/ora_ktbo_poa_s2$ steps/train_mono.sh --nj 1 --cmd run.pl data/train data/lang exp/mono

steps/train_mono.sh --nj 1 --cmd run.pl data/train data/lang exp/mono

steps/train_mono.sh: Initializing monophone system.

run.pl: job failed, log is in exp/mono/log/init.log

manju@dinesh-pc:~/manju/asr/kaldi/my_exp/oracle/ora_ktbo_poa_s2$ 

manju@dinesh-pc:~/manju/asr/kaldi/my_exp/oracle/ora_ktbo_poa_s2$ 

manju@dinesh-pc:~/manju/asr/kaldi/my_exp/oracle/ora_ktbo_poa_s2$ cat exp/mono/log/init.log

# gmm-init-mono --shared-phones=data/lang/phones/sets.int "--train-feats=ark,s,cs:apply-cmvn  --utt2spk=ark:data/train/split1/1/utt2spk scp:data/train/split1/1/cmvn.scp scp:data/train/split1/1/feats.scp ark:- | add-deltas ark:- ark:- | subset-feats --n=10 ark:- ark:-|" data/lang/topo 144 exp/mono/0.mdl exp/mono/tree 

# Started at Sun Mar  5 10:41:03 IST 2017

#

gmm-init-mono --shared-phones=data/lang/phones/sets.int '--train-feats=ark,s,cs:apply-cmvn  --utt2spk=ark:data/train/split1/1/utt2spk scp:data/train/split1/1/cmvn.scp scp:data/train/split1/1/feats.scp ark:- | add-deltas ark:- ark:- | subset-feats --n=10 ark:- ark:-|' data/lang/topo 144 exp/mono/0.mdl exp/mono/tree 

add-deltas ark:- ark:- 

apply-cmvn --utt2spk=ark:data/train/split1/1/utt2spk scp:data/train/split1/1/cmvn.scp scp:data/train/split1/1/feats.scp ark:- 

subset-feats --n=10 ark:- ark:- 

ERROR (gmm-init-mono:main():gmm-init-mono.cc:112) bad variance

[ Stack-Trace: ]

gmm-init-mono() [0x5d2d2e]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*)

kaldi::MessageLogger::~MessageLogger()

main

__libc_start_main

gmm-init-mono() [0x4c8dc9]

# Accounting: time=0 threads=1

# Ended (code 255) at Sun Mar  5 10:41:03 IST 2017, elapsed time 0 seconds

My feature vector has 39 MFCCs (13 MFCCs + delta + delta-delta features) PLUS another 9 values  (which is one among the below 9 lines).

0 0 0 0 0 0 0 0 1

0 0 0 0 0 0 0 1 0

0 0 0 0 0 0 1 0 0

0 0 0 0 0 1 0 0 0

0 0 0 0 1 0 0 0 0

0 0 0 1 0 0 0 0 0

0 0 1 0 0 0 0 0 0

0 1 0 0 0 0 0 0 0

1 0 0 0 0 0 0 0 0

Thanks a lot

[Daniel Povey]

It's trying to do a flat-start initialization of the Gaussians but it needs the variance per dimension.  It looks like for the data  you gave it, the variance for one dimension was exactly zero.  Your features may not be suitable.

Dan

--
Go to http://kaldi-asr.org/forums.html find out how to join
---
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+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[manjunath ke]

Hi Dan

Thanks for the reply.

Yes I have few dimensions having zero values.

I am doing it for some research purpose.

Is there a way to avoid this error and do the experiment.

Thanks

Manjunath

To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.

[Daniel Povey]

You'd have to change the code to apply some floor value in that case.  But it doesn't make sense to build a GMM-based system on features that are either zero or one.

I don't have time to help further.