error from pockectsphinx_batch

[tfpeach]

Hi, dear all,

I am using a trained model and pockectsphinx_batch to recognize one file with the command:

pocketsphinx_batch -hmm model_parameters/UTDallas-concussion.cd_cont_200 -jsgf Debug/CompMod/test.gram \
-dict Dict/UTDallas-concussion.dic -hypseg Debug/CompMod/Timinginfo.txt -cepdir Debug/CompMod/ \
-adcin yes -ctl Debug/CompMod/controlFile.ctl \
-cepext .wav -remove_silence no

However, the result pop up a error:

FATAL: "fsg_lextree.c", line 718: #phones > 128; increase FSG_PNODE_CTXT_BVSZ and recompile

If I remove the option -jsgf, everything will go through well. My grammar file is listed as below:

~~~~~~~~~~~

JSGF V1.0;

grammar dog;
public = <simple>;
<simple>=THE BIG BLACK DOG CAUGHT THE STICK;
~~~~~~~~~

Does anyone know this problem? Thank you very much!

[Nickolay V. Shmyrev]

The message says you your model has too many phones. Use less phones and it will work.

[tfpeach]

Thank you very much for your reply. Yes, I realize this is the reason. However, we are doing about the word dependent phoneme which will have a big phoneme set. I cannot reduce the set, it's given. Is there any way to play with the .c file to rise the threshold?

Thank you once again.

[Nickolay V. Shmyrev]

This is a bad idea. If you have more than 20 words, don't use word-dependent phonemes.

[tfpeach]

Thank you. I have fixed that. Just set the FSG_PNODE_CTXT_BVSZ parameter a little larger. Then I can go through it.

[bd cool]

same thing happene here also, and fixed in the fsg_lextree.h not in fsg_lextree.c