word_align.pl failed with error code 65280 at /usr/local/lib/sphinxtrain/scripts/decode/slave.pl line 173

[ozi samur]

Hi everyone,

I have encountering a problem when I run the "sphinxtrain run".
Actually I have another model which works correctly so we can say that eveything about the binaries, libraries are running correctly.

When I run sphinxtrain run in the "MODULE: 50 Training Context dependent models" module I was getting a few error like this:

**INFO: cmn.c(137): CMN: 57.74  7.13  2.61 12.72  3.61 -0.76 -7.98 -19.27 -12.44 -1.02 -5.23 -1.98  2.09 
ERROR: "backward.c", line 421: Failed to align audio to trancript: final state of the search is not reached
ERROR: "baum_welch.c", line 324: speaker_0001/speaker_0001_0423 ignored**

All of the wav files are 16bit, 16KHZ and mono, I checked it.
All of the number of lines for train fileids and transcription are same and same again for test files.
All of the pronunciation in these file are OK.

Anyway in the decoding part while training I am getting this error:

MODULE: 60 Lattice Generation
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 61 Lattice Pruning
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 62 Lattice Format Conversion
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 65 MMIE Training
Skipped:  $ST::CFG_MMIE set to 'no' in sphinx_train.cfg
MODULE: 90 deleted interpolation
Skipped for continuous models
MODULE: DECODE Decoding using models previously trained
        Decoding 215 segments starting at 0 (part 1 of 1)
        0%
        Aligning results to find error rate
Can't open /Users/xyz/Projects/model1/result/talt_1-1-1.match
word_align.pl failed with error code 65280 at /usr/local/lib/sphinxtrain/scripts/decode/slave.pl line 173.

Is this related to files which status is error in module 50 as I mentioned in above? If I remove the errored files sphinxtrain can find another file. But may be 0.01 of the files have this kind of error not more.

And also I am looking of the log files which is in the logdir/decode. There is just one log file and there is just timestamp is there like below:

Sat Dec  3 14:15:09 2016
        Sat Dec  3 14:15:09 2016

What are your recommendations?

[ozi samur]

By the way at the end of html log file I see this:

MODULE: DECODE Decoding using models previously trained (2016-12-03 14:15)

Decoding 215 segments starting at 0 (part 1 of 1)
pocketsphinx_batch Log File
completedcompleted
Aligning results to find error rate

But I did not set anything about pocketsphinx, I am just using sphinx4.When I try this command:

sphinxtrain -s decode run

I am getting the same error again. Why training gives this error ?

Thanks.

[Arseniy Gorin]

Is this related to files which status is error in module 50 as I mentioned in above?

no, that error can happen sometimes if some words are not well pronounced / audio has a too long silence or smth like that. It is not crucial if you have just few such errors

I am getting the same error again. Why training gives this error ?

Something wrong with decoder. Check logs in logs/decode folder or send it to us.

I am just using sphinx4

This pipeline tries to run pocketsphinx by default. If you install pocketsphinx thigs will be just fine

If you really want to work with sphinx4, you should chekck this tutorial:
http://cmusphinx.sourceforge.net/wiki/tutorialsphinx4

(I did not work much with 4th cersion so it might be the case that someone integrated something handy for 4th version in the training pipeline)

[ozi samur]

Yes, in Mac I could not install the pocketsphinx because of linker problem. When I switch back to Linux, build pocketsphinx and it works.

But in my another model it builds the model without using pocketsphinx, I could see the SER and WER. In that model probably pocketsphinx was not be used. Actually I did not understand exactly pocketsphinx is always being used or not.

Thanks for help.