static int search_cb(search_t *search, search_event_t *evt, void *udata)
{
batch_decoder_t *bd = (batch_decoder_t *) udata;
dict_t *d = search_factory_d2p(bd->sf)->dict;
double delta = get_time_delta(bd);
double frate = cmd_ln_int32_r(search_config(search), "-frate");
FILE *hypfh = NULL;
void *val;
if (hash_table_lookup(bd->hypfiles, search_name(search), &val) == 0)
hypfh = val;
else
hypfh = stdout;
fprintf(hypfh, "time delta %f ", delta);
switch (evt->event) {
case SEARCH_PARTIAL_RESULT: {
int32 score;
seg_iter_t *seg = search_seg_iter(search, &score);
fprintf(hypfh, "partial: ");
for (; seg; seg = seg_iter_next(seg)) {
int sf, ef;
seg_iter_times(seg, &sf, &ef);
fprintf(hypfh, "%s:%.3f ", dict_basestr(d, seg_iter_wid(seg)),
(double) ef / frate);
}
fprintf(hypfh, "(%s)\n", search_uttid(search));
break;
}
case SEARCH_START_UTT:
fprintf(hypfh, "start %s\n", search_uttid(search));
break;
case SEARCH_END_UTT:
fprintf(hypfh, "end %s\n", search_uttid(search));
break;
case SEARCH_FINAL_RESULT: {
int32 score;
seg_iter_t *seg = search_seg_iter(search, &score);
fprintf(hypfh, "full: ");
for (; seg; seg = seg_iter_next(seg)) {
int sf, ef;
seg_iter_times(seg, &sf, &ef);
fprintf(hypfh, "%s:%.3f ", dict_basestr(d, seg_iter_wid(seg)),
(double) ef / frate);
}
fprintf(hypfh, "(%s)\n", search_uttid(search));
break;
}
}
return 0;
}
static int
write_ctm(FILE *fh, ps_decoder_t *ps, ps_seg_t *itor, char const *uttid, int32 frate)
{
logmath_t *lmath = ps_get_logmath(ps);
char *dupid, *show, *channel, *c;
double ustart = 0.0;
/* We have semi-standardized on comma-separated uttids which
* correspond to the fields of the STM file. So if there's a
* comma in the uttid, take the first two fields as show and
* channel, and also try to find the start time. */
show = dupid = ckd_salloc(uttid ? uttid : "(null)");
if ((c = strchr(dupid, ',')) != NULL) {
*c++ = '\0';
channel = c;
if ((c = strchr(c, ',')) != NULL) {
*c++ = '\0';
if ((c = strchr(c, ',')) != NULL) {
ustart = atof_c(c + 1);
}
}
}
else {
channel = NULL;
}
while (itor) {
int32 prob, sf, ef, wid;
char const *w;
/* Skip things that aren't "real words" (FIXME: currently
* requires s3kr3t h34d3rz...) */
w = ps_seg_word(itor);
wid = dict_wordid(ps->dict, w);
if (wid >= 0 && dict_real_word(ps->dict, wid)) {
prob = ps_seg_prob(itor, NULL, NULL, NULL);
ps_seg_frames(itor, &sf, &ef);
fprintf(fh, "%s %s %.2f %.2f %s %.3f\n",
show,
channel ? channel : "1",
ustart + (double)sf / frate,
(double)(ef - sf) / frate,
/* FIXME: More s3kr3tz */
dict_basestr(ps->dict, wid),
logmath_exp(lmath, prob));
}
itor = ps_seg_next(itor);
}
ckd_free(dupid);
return 0;
}
int
dict_altid(dict_t *d, int32 wid)
{
char const *basestr, *wordstr, *c;
if (wid < 0 || wid >= d->n_word)
return 0;
if (wid == dict_basewid(d, wid))
return 1;
/* FIXME: the order in which they appear in nextalt is unrelated
* to the actual alterate indices in the words... */
wordstr = dict_wordstr(d, wid);
basestr = dict_basestr(d, wid);
assert(strlen(wordstr) > strlen(basestr));
c = wordstr + strlen(basestr);
assert(*c == '(');
++c;
return atoi(c);
}
