I can't seem to get the -outlatdir and -nbestdir switches to "work" when
invoked from command line with pocketsphinx_continuous (I'm using v0.6x
still... hopefully that is not the problem).
Are you attempting to obtain the "best" n hypotheses for a given utterance? If
so, I am in need of doing the same thing. I don't want to ask a duplicate
question. I'll monitor this question and can help out (try things on my end)
if need be.
Thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I can't seem to get the -outlatdir and -nbestdir switches to "work" when
invoked from command line with pocketsphinx_continuous (I'm using v0.6x
still... hopefully that is not the problem).
I am invoking using a line like these ones:
pocketsphinx_continuous -hmm ./amodeldir/ -lm ./lmdir/ -dict ./dictdir
-nbestdir ./nbestdir/ -nbest 10
pocketsphinx_continuous -hmm ./amodeldir/ -lm ./lmdir/ -dict ./dictdir
-outlatdir ./lattice
Neither of those create anything at all (both directories already exist).
Are those switches disabled now? Maybe I should ask: can I get not just ONE
recognition "return" but, say, the "best" 3 options the ASR came up with?
Thanks in advance for any help. Will be greatly appreciated, since (as you
surely noted) I am a total noob in these matters.
vitorenes,
Are you attempting to obtain the "best" n hypotheses for a given utterance? If
so, I am in need of doing the same thing. I don't want to ask a duplicate
question. I'll monitor this question and can help out (try things on my end)
if need be.
Thanks.
That's exactly it. Any help will be greatly appreciated - I'll post any
findings here as well.
Cheers.
They are not supposed to. -nbestdir is an option for pocketsphinx_batch, not
pocketsphinx_continuous
Yes, you can, you can write your own tool using pocketsphinx API.
Look at pocketsphinx.c :
for (seg = ps_seg_iter(ps, &score); seg;
seg = ps_seg_next(seg)) {
char const *word;
int sf, ef;
int32 post, lscr, ascr, lback;
word = ps_seg_word(seg);
ps_seg_frames(seg, &sf, &ef);
post = ps_seg_prob(seg, &ascr, &lscr, &lback);
E_INFO_NOFN("%-20s %-5d %-5d %-1.3f %-10d %-10d %-3d\n",
word, sf, ef, logmath_exp(ps_get_logmath(ps), post), ascr, lscr, lback);
}
that's almost done ;)