iimproving results of recognizing

[osm slm]

I used sphinx3 to recognize some discrete words
and I used lmtool to generate language model and dictionary and this acoustic
model
"4000 senone, 64 Gaussian continuous density models" for (8kHz) telephone
speech
but the recognizing results is poor and slow
I know that this acoustic model is very large but what 's the alternative? can
I use semi-continuous model with sphinx3? if I can, how?
and if I used pocketsphinx? will it be faster? and can I use it for batch mode
recognizing?
cause I want this recognizing to be in real time and as accurate as possible

[Nickolay V. Shmyrev]

but the recognizing results is poor and slow

I think your recognition results are fast and accurate, try to look
differently on them

can I use semi-continuous model with sphinx3?

Yes

If I can, how?

add

-senmgau .semi.

and if I used pocketsphinx? will it be faster?

yes

and can I use it for batch mode recognizing?

yes