int
mk_phone_seq(acmod_id_t **out_phone,
uint32 *out_n_phone,
char *trans,
acmod_set_t *acmod_set,
lexicon_t *lex)
{
char **word;
uint32 n_word;
uint32 n_phone;
acmod_id_t *phone;
char *btw_mark;
word = mk_wordlist(trans, &n_word);
phone = mk_phone_list(&btw_mark, &n_phone, word, n_word, lex);
if (phone == NULL) {
E_WARN("Unable to produce phonetic transcription for the utterance '%s'.\n", corpus_utt());
}
cvt2triphone(acmod_set, phone, btw_mark, n_phone);
ckd_free(btw_mark);
ckd_free(word);
*out_phone = phone;
*out_n_phone = n_phone;
return S3_SUCCESS;
}
state_t *next_utt_states(uint32 *n_state,
lexicon_t *lex,
model_inventory_t *inv,
model_def_t *mdef,
char *trans,
int32 sil_del,
char* silence_str
)
{
char **word;
uint32 n_word;
uint32 n_phone;
char *btw_mark;
acmod_set_t *acmod_set;
acmod_id_t *phone;
acmod_id_t optSil;
state_t *state_seq;
word = mk_wordlist(trans, &n_word);
phone = mk_phone_list(&btw_mark, &n_phone, word, n_word, lex);
if (phone == NULL) {
E_WARN("Unable to produce CI phones for utt\n");
ckd_free(word);
return NULL;
}
acmod_set = inv->acmod_set;
#ifdef NEXT_UTT_STATES_VERBOSE
print_phone_list(phone, n_phone, btw_mark, acmod_set);
#endif
cvt2triphone(acmod_set, phone, btw_mark, n_phone);
#ifdef NEXT_UTT_STATES_VERBOSE
print_phone_list(phone, n_phone, btw_mark, acmod_set);
#endif
optSil= acmod_set_name2id(acmod_set, silence_str);
/*
* Debug?
* E_INFO("Silence id %d\n",optSil);
*/
state_seq = state_seq_make(n_state, phone, n_phone, inv, mdef,sil_del,(acmod_id_t)optSil);
#ifdef NEXT_UTT_STATES_VERBOSE
state_seq_print(state_seq, *n_state, mdef);
#endif
ckd_free(phone);
ckd_free(btw_mark);
ckd_free(word);
return state_seq;
}
int
agg_phn_seg(lexicon_t *lex,
acmod_set_t *acmod_set,
feat_t *fcb,
segdmp_type_t type)
{
uint16 *seg;
vector_t *mfcc;
vector_t **feat;
uint32 n_frame;
uint32 tick_cnt;
acmod_id_t *phone;
uint32 *start;
uint32 *len;
uint32 n_phone;
uint32 s;
char *btw_mark;
char *trans;
char **word;
uint32 n_word;
int32 mfc_veclen = cmd_ln_int32("-ceplen");
uint32 n_stream;
const uint32 *veclen;
tick_cnt = 0;
n_stream = feat_dimension1(fcb);
veclen = feat_stream_lengths(fcb);
while (corpus_next_utt()) {
if ((++tick_cnt % 500) == 0) {
printf("[%u] ", tick_cnt);
fflush(stdout);
}
if (corpus_get_sent(&trans) != S3_SUCCESS) {
E_FATAL("Unable to read word transcript for %s\n", corpus_utt_brief_name());
}
if (corpus_get_seg(&seg, &n_frame) != S3_SUCCESS) {
E_FATAL("Unable to read Viterbi state segmentation for %s\n", corpus_utt_brief_name());
}
word = mk_wordlist(trans, &n_word);
phone = mk_phone_list(&btw_mark, &n_phone, word, n_word, lex);
start = ckd_calloc(n_phone, sizeof(uint32));
len = ckd_calloc(n_phone, sizeof(uint32));
/* check to see whether the word transcript and dictionary entries
agree with the state segmentation */
if (ck_seg(acmod_set, phone, n_phone, seg, n_frame, corpus_utt()) != S3_SUCCESS) {
free(trans); /* alloc'ed using strdup, not ckd_*() */
free(seg); /* alloc'ed using malloc in areadshort(), not ckd_*() */
ckd_free(word);
ckd_free(phone);
E_ERROR("ck_seg failed");
continue;
}
if (cvt2triphone(acmod_set, phone, btw_mark, n_phone) != S3_SUCCESS) {
free(trans); /* alloc'ed using strdup, not ckd_*() */
free(seg); /* alloc'ed using malloc in areadshort(), not ckd_*() */
ckd_free(word);
ckd_free(phone);
E_ERROR("cvt2triphone failed");
continue;
}
ckd_free(btw_mark);
if (mk_seg(acmod_set,
seg,
n_frame,
phone,
start,
len,
n_phone) != S3_SUCCESS) {
free(trans);
free(seg);
ckd_free(word);
ckd_free(phone);
E_ERROR("mk_seg failed");
continue;
}
if (corpus_provides_mfcc()) {
if (corpus_get_generic_featurevec(&mfcc, &n_frame, mfc_veclen) < 0) {
E_FATAL("Can't read input features from %s\n", corpus_utt());
}
if (n_frame < 9) {
E_WARN("utt %s too short\n", corpus_utt());
if (mfcc) {
ckd_free(mfcc[0]);
ckd_free(mfcc);
mfcc = NULL;
}
continue;
}
feat = feat_array_alloc(fcb, n_frame + feat_window_size(fcb));
feat_s2mfc2feat_live(fcb, mfcc, &n_frame, TRUE, TRUE, feat);
for (s = 0; s < n_phone; s++) {
segdmp_add_feat(phone[s],
&feat[start[s]],
len[s]);
}
feat_array_free(feat);
free(&mfcc[0][0]);
ckd_free(mfcc);
}
else {
E_FATAL("No data type specified\n");
}
free(trans); /* alloc'ed using strdup, not ckd_*() */
free(seg); /* alloc'ed using malloc in areadshort(), not ckd_*() */
ckd_free(word);
ckd_free(phone);
ckd_free(start);
ckd_free(len);
}
return 0;
}
