int default_utt_break(cst_tokenstream *ts,
const char *token,
cst_relation *tokens)
{
/* This is the default utt break functions, languages may override this */
/* This will be ok for some latin based languages */
const char *postpunct = item_feat_string(relation_tail(tokens), "punc");
const char *ltoken = item_name(relation_tail(tokens));
if (cst_strchr(ts->whitespace,'\n') != cst_strrchr(ts->whitespace,'\n'))
/* contains two new lines */
return TRUE;
else if (strchr(postpunct,':') ||
strchr(postpunct,'?') ||
strchr(postpunct,'!'))
return TRUE;
else if (strchr(postpunct,'.') &&
(cst_strlen(ts->whitespace) > 1) &&
strchr("ABCDEFGHIJKLMNOPQRSTUVWXYZ",token[0]))
return TRUE;
else if (strchr(postpunct,'.') &&
/* next word starts with a capital */
strchr("ABCDEFGHIJKLMNOPQRSTUVWXYZ",token[0]) &&
/* last word isn't an abbreviation */
!(strchr("ABCDEFGHIJKLMNOPQRSTUVWXYZ",ltoken[cst_strlen(ltoken)-1])||
((cst_strlen(ltoken) < 4) &&
strchr("ABCDEFGHIJKLMNOPQRSTUVWXYZ",ltoken[0]))))
return TRUE;
else
return FALSE;
}
/* Dummy F0 modelling for phones, copied directly from us_f0_model.c */
cst_utterance *flat_prosody(cst_utterance *u)
{
/* F0 target model */
cst_item *s,*t;
cst_relation *targ_rel;
float mean, stddev;
targ_rel = utt_relation_create(u,"Target");
mean = get_param_float(u->features,"target_f0_mean", 100.0);
mean *= get_param_float(u->features,"f0_shift", 1.0);
stddev = get_param_float(u->features,"target_f0_stddev", 12.0);
s=relation_head(utt_relation(u,"Segment"));
t = relation_append(targ_rel,NULL);
item_set_float(t,"pos",0.0);
item_set_float(t,"f0",mean+stddev);
s=relation_tail(utt_relation(u,"Segment"));
t = relation_append(targ_rel,NULL);
item_set_float(t,"pos",item_feat_float(s,"end"));
item_set_float(t,"f0",mean-stddev);
return u;
}
cst_utterance *russian_postlex_function(cst_utterance *u)
{
const cst_item *word,*seg;
const char *answer,*name,*pair;
for(word=relation_head(utt_relation(u,"Transcription"));word;word=item_next(word))
{
if(item_feat_present(word,"no_pl")||item_feat_present(word,"no_vr"))
continue;
for(seg=item_daughter(word);seg;seg=item_next(seg))
{
name=item_feat_string(seg,"name");
if(cst_member_string(name,unstressed_vowels))
{
answer=val_string(cart_interpret(item_as(seg,"Segment"),&ru_vowel_reduction_cart));
if(!cst_streq(answer,"N"))
item_set_string(seg,"name",answer);
}
else
{
if(cst_streq(name,"ii")&&
cst_streq(ffeature_string(seg,"R:Segment.p.ph_csoft"),"-")&&
!(cst_streq(item_feat_string(word,"name"),"и")&&
cst_streq(ffeature_string(word,"gpos"),"content")))
{
item_set_string(seg,"name","yy");
}
}
}
}
for(word=relation_tail(utt_relation(u,"Transcription"));word;word=item_prev(word))
{
if(item_feat_present(word,"no_pl"))
continue;
for(seg=item_last_daughter(word);seg;seg=item_prev(seg))
{
name=item_feat_string(seg,"name");
pair=russian_vpair(name);
if(pair!=NULL)
{
answer=val_string(cart_interpret(item_as(seg,"Segment"),&ru_vpair_cart));
if(cst_streq(answer,"Y"))
item_set_string(seg,"name",pair);
}
}
}
return u;
}
