another nbest-related question

[chris]

Hi all,

I'm experimenting with nbest lists for my fsg with around 1500 entries in a
flat list.

This is the code I use to retrieve the list:

    nbest_iter = ps_nbest(ps,0, -1, NULL, NULL);
    int i;
    n = 3;
    for (i = 0; i < n && nbest_iter && (nbest_iter = ps_nbest_next(nbest_iter)); i++) {
        nbest_hyp = ps_nbest_hyp(nbest_iter, &path_score);

        if (nbest_hyp != NULL){
            printf("Nbest: %s with path_score %d\n", nbest_hyp, path_score);
        }
    }

What I get as a result, however, is three times the same hypothesis with the
same score. If I increase n to 100, I get the same hypothesis a hundred times,
but at least I see 3-4 different probabilities.

My understanding of the nbest list up to now was that it is a list of
differing hypotheses. I'm not quite sure now whether I'm doing something wrong
here or whether thats's the way the nbest list looks like in pocketsphinx.

Any thoughts on this are appreciated.

Regards,
Chris

[Nickolay V. Shmyrev]

This behaviour is expected.

[chris]

hmm. is it because of the fsg? is it different when you use SLMs?
'cause it doesn't seem too useful to me, at first glance. A Nuance ASR, for
instance, would almost certainly have a couple of different hypotheses when
given the same grammar.

c.

[Nickolay V. Shmyrev]

The function to join identical n-best hypothesis is not implemented,

[chris]

I see. Thanks for the info!

c.