Using pocketsphinx compiled from the trunk, I sometimes get hypothesis that are not allowed by the grammar. For example, using RM1 acoustic model, with:
I get:
WHERE DID HE GO BEN GO (recording -53936196)
which is not in my grammar (BEN is at the final state, nothing can be recognized afterwards).
The recording and the files dict, ctl and fsg are available there: http://sylvainchevalier.free.fr/package.tgz
I did not have problems with pocketsphinx 0.5 and sphinx 3. It looks like a bug in pocketsphinx (or sphinxbase) to me. Any clue?
Thanks a lot!
Sylvain
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I never got any problems of hyp not allowed by the grammar using sphinx 3. So, it should be a problem of pocketsphinx, not sphinxbase. Do you know if there is a way to retrieve the (fsg) states sequence ?
Thanks!
Sylvain
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
And, the strange think is that I tried the same with pocketsphinx-0.5 and it did the same. Sphinx3 returns better WHERE DID HE GO but it's a different decoder.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
Using pocketsphinx compiled from the trunk, I sometimes get hypothesis that are not allowed by the grammar. For example, using RM1 acoustic model, with:
/data/sylvain/cmusphinx/trunk/pocketsphinx/src/programs/pocketsphinx_batch -dict dict -ctl ctl -fsg fsg -cepdir ./ -hmm /data/sylvain/cmusphinx/trunk/sphinx3/model/hmm/RM1_cd_semi -feat s2_4x -fdict /data/sylvain/cmusphinx/trunk/sphinx3/model/lm/an4/filler.dict -hyp hyp -adchdr 44 -adcin yes -cepext .wav
I get:
WHERE DID HE GO BEN GO (recording -53936196)
which is not in my grammar (BEN is at the final state, nothing can be recognized afterwards).
The recording and the files dict, ctl and fsg are available there: http://sylvainchevalier.free.fr/package.tgz
I did not have problems with pocketsphinx 0.5 and sphinx 3. It looks like a bug in pocketsphinx (or sphinxbase) to me. Any clue?
Thanks a lot!
Sylvain
Many thanks Nickolay for looking into it.
I never got any problems of hyp not allowed by the grammar using sphinx 3. So, it should be a problem of pocketsphinx, not sphinxbase. Do you know if there is a way to retrieve the (fsg) states sequence ?
Thanks!
Sylvain
Y, it might be pocketsphinx issue really. We need to look on this closer, but no ready solution yet.
Hey Sylvain
Probably I'm asking about too much, but can you just bisect and find the exact revision where regression happens
And, the strange think is that I tried the same with pocketsphinx-0.5 and it did the same. Sphinx3 returns better WHERE DID HE GO but it's a different decoder.