Maybe it is bug or Not in BW

[LEI YANG]

Dear Sir,

I meet following question about how to correspond the gauden_t in Sphinx, Can you tell me the relationship between state_t.cb with g->mean, I found there are some problem in gauden.cpp . When I run log_full_densities, program dead because g->mean[mgau]=0x00000000, I think the reason is mgau is too large or it is wrong code.

The case is Input words is “FINISH”, and state_seq[0].cb =17534, and mean is n_mgau = 315, n_feat = 4, n_density = 1, vlen is 12  total size of mean in gauden_t is =15120. Can you help me analysis it ?

int gauden_compute_log(float64 **den,        /* density array for a mixture Gaussian */
                       uint32 **den_idx,    /* density index array for n_top < n_density eval */
                       vector_t *obs,        /* observation vector for some time */
                       gauden_t *g,            /* Gaussian density structure */
                       uint32 mgau)            /* id of the mixture Gau. to evaluate */
{
&#8230;
    if (g->n_top == g->n_density)
    {
        for (j = 0; j < g->n_feat; j++)
        {
            log_full_densities(den[j],den_idx[j],g->n_density,g->veclen[j],obs[j],
                               g->mean[mgau][j],g->var[mgau][j],g->norm[mgau][j]);
        }
    }
    else
    {
        assert(g->n_top < g->n_density);

        for (j = 0; j < g->n_feat; j++)
        {
            log_topn_densities(den[j],den_idx[j],g->n_top,g->n_density,g->veclen[j],
                               obs[j],g->mean[mgau][j],g->var[mgau][j],g->norm[mgau][j]);

        }
    }
&#8230;

    return S3_SUCCESS;
}

[Akkineni Rajesh]

you said n_density = 1...

check it again..i think it may be 256 by default..