static int ifd_syl_boundary(const cst_item *i,const cst_val *v)
{
/* Returns TRUE if this should be a syllable boundary */
/* This is of course phone set dependent */
int p, n, nn;
if (v == NULL)
return TRUE;
else if (ifd_is_silence(val_string(val_car(v))))
return TRUE;
else if (!ifd_has_vowel_in_list(v)) /* no more vowels so rest *all* coda */
return FALSE;
else if (!ifd_has_vowel_in_syl(i)) /* need a vowel */
return FALSE;
else if (ifd_is_vowel(val_string(val_car(v))))
return TRUE;
else if (val_cdr(v) == NULL)
return FALSE;
else
{ /* so there is following vowel, and multiple phones left */
p = ifd_sonority(item_feat_string(i,"name"));
n = ifd_sonority(val_string(val_car(v)));
nn = ifd_sonority(val_string(val_car(val_cdr(v))));
if ((p <= n) && (n <= nn))
return TRUE;
else
return FALSE;
}
}
cst_val *lts_apply_val(const cst_val *wlist,const char *feats,const cst_lts_rules *r)
{
/* for symbol to symbol mapping */
const cst_val *v;
cst_val *p;
char *word;
int i,j;
word = cst_alloc(char,val_length(wlist)+1);
for (v=wlist,i=0; v; v=val_cdr(v),i++)
{
for (j=0; r->letter_table[j]; j++)
if (cst_streq(val_string(val_car(v)),r->letter_table[j]))
{
word[i] = j;
break;
}
if (!r->letter_table[j])
{
#if 0
printf("awb_debug unknown letter >%s<\n",val_string(val_car(v)));
#endif
i--; /* can't find this letter so skip it */
}
}
p = lts_apply(word,feats,r);
cst_free(word);
return p;
}
int main(int argc, char **argv)
{
cst_wave *nw, *all;
cst_val *files;
const cst_val *w;
cst_val *wavelist;
cst_features *args;
int i,j;
float ntime;
int stime;
const char *nwfile;
args = new_features();
files =
cst_args(argv,argc,
"usage: combine_waves OPTIONS\n"
"Combine waves into single waveform\n"
"-o <string> Output waveform\n"
"-f <int> Input sample rate (for raw input)\n"
"-itype <string> Input type, raw or headered\n"
"-wavelist <string> File containing times and wave filenames\n",
args);
wavelist = get_wavelist(get_param_string(args,"-wavelist","-"));
if (wavelist == 0)
return -1;
all = new_wave();
for (w = wavelist; w; w = val_cdr(w))
{
ntime = decode_time(val_string(val_car(w)));
nwfile = val_string(val_car(val_cdr(w)));
nw = new_wave();
if (cst_wave_load_riff(nw,nwfile) != CST_OK_FORMAT)
{
fprintf(stderr,
"combine_waves: can't read file or wrong format \"%s\"\n",
nwfile);
continue;
}
stime = ntime * nw->sample_rate;
cst_wave_resize(all,stime+nw->num_samples,1);
for (i=0,j=stime; i<nw->num_samples; i++,j++)
{
/* this will cause overflows */
all->samples[j] += nw->samples[i];
}
delete_wave(nw);
}
cst_wave_save_riff(all,get_param_string(args,"-o","-"));
return 0;
}
int cmu_grapheme_syl_boundary(const cst_item *i,const cst_val *rest)
{
if (!rest)
return TRUE;
else if (!cmu_grapheme_contains_vowel(rest))
return FALSE;
else if (!cmu_grapheme_has_vowel_in_syl(i))
return FALSE;
#if 0
else if (rest && val_cdr(rest) &&
cst_streq("n",val_string(val_car(rest))) &&
!cmu_grapheme_is_vowel(val_string(val_car(rest))))
return FALSE;
else if (rest && val_cdr(rest) &&
cmu_grapheme_is_vowel(ffeature_string(i,"name")) &&
!cmu_grapheme_is_vowel(val_string(val_car(rest))) &&
!cmu_grapheme_is_vowel(val_string(val_car(val_cdr(rest)))))
return FALSE;
else if (rest && val_cdr(rest) && val_cdr(val_cdr(rest)) &&
!cmu_grapheme_is_vowel(val_string(val_car(rest))) &&
!cmu_grapheme_is_vowel(val_string(val_car(val_cdr(rest)))) &&
!cmu_grapheme_is_vowel(val_string(val_car(val_cdr(val_cdr(rest))))))
return FALSE;
else if (rest && val_cdr(rest) &&
(cst_streq(val_string(val_car(rest)),
val_string(val_car(val_cdr(rest))))))
return FALSE;
#endif
else
return TRUE;
}
const cst_val *val_assoc_string(const char *v1,const cst_val *al)
{
const cst_val *i;
for (i=al; i; i=val_cdr(i))
{
if (cst_streq(v1,val_string(val_car(val_car(i)))))
return val_car(i);
}
return NULL;
}
static int cmulex_onset_bigram(const cst_val *rest)
{
char x[10];
int i;
cst_sprintf(x,"%s%s",val_string(val_car(rest)),
val_string(val_car(val_cdr(rest))));
for (i=0; cmulex_onset_bigrams[i]; i++)
if (cst_streq(x,cmulex_onset_bigrams[i]))
return TRUE;
return FALSE;
}
static int context_match(const cst_val *PATTERN, const cst_val *STRING,
const cst_val *sets)
{
if (!PATTERN)
return TRUE;
else if (!STRING)
return FALSE;
else if (val_cdr(PATTERN) &&
(cst_streq("*",val_string(val_car(PATTERN)))))
return
context_match(val_cdr(val_cdr(PATTERN)),STRING,sets) || /* skip */
context_match(val_cdr(PATTERN),STRING,sets) || /* last match */
(item_match(val_car(val_cdr(PATTERN)),val_car(STRING),sets) &&
context_match(val_cdr(val_cdr(PATTERN)), /* loop match */
val_cdr(STRING),sets));
#if 0
else if (val_cdr(PATTERN) &&
(cst_streq("+",val_string(val_car(PATTERN)))))
return context_match(val_cdr(PATTERN),STRING,sets) || /* last match */
(item_match(val_car(val_cdr(PATTERN)),val_car(STRING),sets) &&
context_match(val_cdr(val_cdr(PATTERN)), /* loop match */
val_cdr(STRING),sets));
#endif
else if (item_match(val_car(PATTERN),val_car(STRING),sets))
return context_match(val_cdr(PATTERN),val_cdr(STRING),sets);
else
return FALSE;
}
cst_val *en_exp_ordinal(const char *rawnumstring)
{
/* return ordinal for digit string */
cst_val *card, *o;
const cst_val *t;
const char *l;
const char *ord;
char *numstring;
int i,j;
numstring = cst_strdup(rawnumstring);
for (j=i=0; i < cst_strlen(rawnumstring); i++)
if (rawnumstring[i] != ',')
{
numstring[j] = rawnumstring[i];
j++;
}
numstring[j] = '\0';
card = val_reverse(en_exp_number(numstring));
cst_free(numstring);
l = val_string(val_car(card));
ord = 0;
for (i=0; i<10; i++)
if (cst_streq(l,digit2num[i]))
ord = ord2num[i];
if (!ord)
for (i=0; i<10; i++)
if (cst_streq(l,digit2teen[i]))
ord = ord2teen[i];
if (!ord)
for (i=0; i<10; i++)
if (cst_streq(l,digit2enty[i]))
ord = ord2enty[i];
if (cst_streq(l,"hundred"))
ord = "hundredth";
if (cst_streq(l,"thousand"))
ord = "thousandth";
if (cst_streq(l,"billion"))
ord = "billtionth";
if (!ord) /* dunno, so don't convert anything */
return card;
o = cons_val(string_val(ord),0);
for (t=val_cdr(card); t; t=val_cdr(t))
o = cons_val(val_car(t),o);
delete_val(card);
return o;
}
static int WordSylSeg(cst_utterance *u)
{
cst_item *word;
cst_relation *sylstructure,*seg,*syl;
cst_val *phones;
const cst_val *p;
cst_item *ssword,*segitem;
syl = utt_relation_create(u,"Syllable");
sylstructure = utt_relation_create(u,"SylStructure");
seg = utt_relation_create(u,"Segment");
for (word=relation_head(utt_relation(u,"Word"));
word; word=item_next(word))
{
printf("word: %s\n",item_feat_string(word,"name"));
ssword = relation_append(sylstructure,word);
phones = lex_lookup((cst_lexicon *)&cmu_lex,item_feat_string(word,"name"),0);
for (p=phones; p; p=val_cdr(p))
{
segitem = relation_append(seg,NULL);
item_set(segitem,"name",val_car(p));
printf("seg: %s\n",item_feat_string(segitem,"name"));
item_add_daughter(ssword,segitem);
}
delete_val_list(phones);
}
return TRUE;
}
void val_print(cst_file fd,const cst_val *v)
{
const cst_val *p;
if (v == NULL)
cst_fprintf(fd,"[null]");
else if (CST_VAL_TYPE(v) == CST_VAL_TYPE_INT)
cst_fprintf(fd,"%d",val_int(v));
else if (CST_VAL_TYPE(v) == CST_VAL_TYPE_FLOAT)
cst_fprintf(fd,"%f",val_float(v));
else if (CST_VAL_TYPE(v) == CST_VAL_TYPE_STRING)
cst_fprintf(fd,"%s",val_string(v));
else if (cst_val_consp(v))
{
cst_fprintf(fd,"(");
for (p=v; p; )
{
val_print(fd,val_car(p));
p=val_cdr(p);
if (p)
cst_fprintf(fd," ");
}
cst_fprintf(fd,")");
}
else
cst_fprintf(fd,"[Val %s 0x%p]",
cst_val_defs[CST_VAL_TYPE(v)/2].name,CST_VAL_VOID(v));
}
static const cst_val *find_rewrite_rule(const cst_val *LC,
const cst_val *RC,
const cst_lts_rewrites *r)
{
/* Search through rewrite rules to find matching one */
const cst_val *i, *RLC, *RA, *RRC;
for (i=r->rules; i; i=val_cdr(i))
{
/* val_print(stdout, val_car(i)); printf("\n"); */
RLC = val_car(val_car(i));
RA = val_car(val_cdr(val_car(i)));
RRC = val_car(val_cdr(val_cdr(val_car(i))));
if (rule_matches(LC,RC,RLC,RA,RRC,r->sets))
return val_car(i);
}
fprintf(stderr,"LTS_REWRITES: unable to find a matching rules for:\n");
fprintf(stderr,"CL: ");
val_print(stderr,LC);
fprintf(stderr,"\n");
fprintf(stderr,"RC: ");
val_print(stderr,RC);
fprintf(stderr,"\n");
return NULL;
}
static int ifd_has_vowel_in_list(const cst_val *v)
{
const cst_val *t;
for (t=v; t; t=val_cdr(t))
if (ifd_is_vowel(val_string(val_car(v))))
return TRUE;
return FALSE;
}
static int cmulex_dist_to_vowel(const cst_val *rest)
{
if (rest == 0)
return 0; /* shouldn't get here */
else if (cmu_is_vowel(val_string(val_car(rest))))
return 0;
else
return 1+cmulex_dist_to_vowel(val_cdr(rest));
}
int val_member_string(const char *v1,const cst_val *l)
{
const cst_val *i;
for (i=l; i; i=val_cdr(i))
{
if (cst_streq(v1,val_string(val_car(i))))
return TRUE;
}
return FALSE;
}
int val_member(const cst_val *v1,const cst_val *l)
{
const cst_val *i;
for (i=l; i; i=val_cdr(i))
{
if (val_equal(val_car(i),v1))
return TRUE;
}
return FALSE;
}
cst_voice *flite_voice_select(const char *name)
{
const cst_val *v;
cst_voice *voice;
if (flite_voice_list == NULL)
return NULL; /* oops, not good */
if (name == NULL)
return val_voice(val_car(flite_voice_list));
for (v=flite_voice_list; v; v=val_cdr(v))
{
voice = val_voice(val_car(v));
if (cst_streq(name,voice->name))
return voice;
}
return flite_voice_select(NULL);
}
static int cmu_grapheme_contains_vowel(const cst_val *r)
{
const cst_val *x;
for (x=r; x; x=val_cdr(x))
{
if (cmu_grapheme_is_vowel(val_string(val_car(x))))
return TRUE;
}
return FALSE;
}
cst_val *cmu_grapheme_lex_lts_function(const struct lexicon_struct *l,
const char *word, const char *pos,
const cst_features *feats)
{
cst_val *phones = 0;
cst_val *utflets = 0;
const cst_val *v;
char ord[10];
int i,phindex;
/* string to utf8 chars */
utflets = cst_utf8_explode(word);
for (v=utflets; v; v=val_cdr(v))
{
/* We will add the found phones in reverse order and reverse then */
/* afterwards */
cst_utf8_as_hex(val_string(val_car(v)),ord);
phindex = cst_find_u2sampa(ord);
if (phindex < 0)
printf("awb_debug no sampa %s %s\n",val_string(val_car(v)),ord);
for (i=4; (phindex>=0) && (i>0); i--)
{
if (unicode_sampa_mapping[phindex][i])
phones = cons_val(string_val(unicode_sampa_mapping[phindex][i]),
phones);
}
}
phones = val_reverse(phones);
#if 1
printf("cmu_grapheme_lex.c: word \"%s\" ",word);
val_print(stdout,phones);
printf("\n");
#endif
delete_val(utflets);
return phones;
}
static cst_val *add_lts_boundary_marks(const cst_val *l)
{
cst_val *l1;
const cst_val *v;
l1 = cons_val(string_val("#"),NULL);
for (v=l;v;v=val_cdr(v))
{
l1=cons_val(val_car(v),l1);
}
l1 = cons_val(string_val("#"),l1);
l1 = val_reverse(l1);
return l1;
}
static int rule_matches(const cst_val *LC, const cst_val *RC,
const cst_val *RLC, const cst_val *RA,
const cst_val *RRC,
const cst_val *sets)
{
const cst_val *rc, *ra;
/* Check [ X ] bit */
for (rc=RC,ra=RA; ra; ra=val_cdr(ra),rc=val_cdr(rc))
{
if (!rc) return FALSE;
if (!cst_streq(val_string(val_car(ra)),
val_string(val_car(rc))))
return FALSE;
}
/* Check LC bit: LC may have some limited regex stuff */
if (context_match(RLC,LC,sets) && context_match(RRC,rc,sets))
return TRUE;
else
return FALSE;
}
cst_string *cst_implode(const cst_val *sl)
{
const cst_val *v;
int l=0;
char *s;
for (v=sl; v; v=val_cdr(v))
{
if (val_stringp(val_car(v)))
l += cst_strlen(val_string(val_car(v)));
}
s = cst_alloc(cst_string,l+1);
for (v=sl; v; v=val_cdr(v))
{
if (val_stringp(val_car(v)))
cst_sprintf(s,"%s%s",s,val_string(val_car(v)));
}
return s;
}
int val_equal(const cst_val *v1, const cst_val *v2)
{
if (v1 == v2)
return TRUE; /* its eq so its equal */
else if (v1 == 0)
return FALSE;
else if (CST_VAL_TYPE(v1) == CST_VAL_TYPE(v2))
{
if (cst_val_consp(v1))
return ((val_equal(val_car(v1),val_car(v2))) &&
(val_equal(val_cdr(v1),val_cdr(v2))));
else if (CST_VAL_TYPE(v1) == CST_VAL_TYPE_INT)
return (val_int(v1) == val_int(v2));
else if (CST_VAL_TYPE(v1) == CST_VAL_TYPE_FLOAT)
return (val_float(v1) == val_float(v2));
else if (CST_VAL_TYPE(v1) == CST_VAL_TYPE_STRING)
return (cst_streq(CST_VAL_STRING(v1),CST_VAL_STRING(v2)));
else
return CST_VAL_VOID(v1) == CST_VAL_VOID(v2);
}
else
return FALSE;
}
cst_val *lts_rewrites(const cst_val *itape, const cst_lts_rewrites *r)
{
/* Returns list of rewritten "letters" to "phones" by r */
cst_val *LC;
const cst_val *RC, *i;
const cst_val *rule;
cst_val *otape;
LC = cons_val(val_car(itape),NULL);
RC = val_cdr(itape);
otape = NULL;
while (val_cdr(RC))
{
rule = find_rewrite_rule(LC,RC,r);
if (!rule)
break;
/* val_print(stdout,rule);
printf("\n"); */
/* Shift itape head */
for (i=val_car(val_cdr(rule)); i; i=val_cdr(i))
{
LC = cons_val(val_car(RC),LC);
RC = val_cdr(RC);
}
/* Output things to otape */
for (i=val_car(val_cdr(val_cdr(val_cdr(rule)))); i; i=val_cdr(i))
otape = cons_val(val_car(i),otape);
}
delete_val_list(LC);
return val_reverse(otape);
}
int cmu_syl_boundary_mo(const cst_item *i,const cst_val *rest)
{
/* syl boundary maximal onset */
int d2v;
if (rest == NULL)
return TRUE;
else if (cmu_is_silence(val_string(val_car(rest))))
return TRUE;
else if (!cmu_has_vowel_in_list(rest))
/* no more vowels so rest *all* coda */
return FALSE;
else if (!cmu_has_vowel_in_syl(i)) /* need a vowel */
/* no vowel yet in syl so keep copying */
return FALSE;
else if (cmu_is_vowel(val_string(val_car(rest))))
/* next is a vowel, syl has vowel, so this is a break */
return TRUE;
else if (cst_streq("ng",val_string(val_car(rest))))
/* next is "ng" which can't start a word internal syl */
return FALSE;
else
{
/* want to know if from rest to the next vowel is a valid onset */
d2v = cmulex_dist_to_vowel(rest);
if (d2v < 2)
return TRUE;
else if (d2v > 3)
return FALSE;
else if (d2v == 2)
return cmulex_onset_bigram(rest);
else /* if (d2v == 3) */
return cmulex_onset_trigram(rest);
return TRUE;
}
}
static int ru_syl_boundary(const cst_item *i,const cst_val *v)
{
const char *p;
const char *n;
const char *nn;
if (v == NULL)
return TRUE;
n=val_string(val_car(v));
if (is_silence(n))
return TRUE;
if (!has_vowel_in_list(v))
return FALSE;
if (!has_vowel_in_syl(i))
return FALSE;
if (is_vowel(n))
return TRUE;
if (val_cdr(v) == NULL)
return FALSE;
p = item_feat_string(i,"name");
nn = val_string(val_car(val_cdr(v)));
if(is_vowel(p))
{
if(is_vowel(nn))
return TRUE;
if(is_sonorant(n)&&!is_sonorant(nn))
return FALSE;
if(is_j(n))
return FALSE;
return TRUE;
}
if(is_sonorant(p))
return TRUE;
if(is_j(p))
return TRUE;
return FALSE;
}
cst_utterance *default_textanalysis(cst_utterance *u)
{
cst_item *t,*word;
cst_relation *word_rel;
cst_val *words;
const cst_val *w;
const cst_val *ttwv;
word_rel = utt_relation_create(u,"Word");
ttwv = feat_val(u->features, "tokentowords_func");
for (t=relation_head(utt_relation(u,"Token")); t; t=item_next(t))
{
if (ttwv)
words = (cst_val *)(*val_itemfunc(ttwv))(t);
else
words = default_tokentowords(t);
for (w=words; w; w=val_cdr(w))
{
word = item_add_daughter(t,NULL);
if (cst_val_consp(val_car(w)))
{ /* Has extra features */
item_set_string(word,"name",val_string(val_car(val_car(w))));
feat_copy_into(val_features(val_cdr(val_car(w))),
item_feats(word));
}
else
item_set_string(word,"name",val_string(val_car(w)));
relation_append(word_rel,word);
}
delete_val(words);
}
return u;
}
static void flite_voice_list_print(void)
{
cst_voice *voice;
const cst_val *v;
printf("Voices available: ");
for (v=flite_voice_list; v; v=val_cdr(v))
{
voice = val_voice(val_car(v));
printf("%s ",voice->name);
}
printf("\n");
return;
}
char *ru_implode(const cst_val *l)
{
const cst_val *v;
size_t n=1;
char *s,*p;
for(v=l;(v!=NULL);v=val_cdr(v))
{
if(val_stringp(val_car(v)))
n+=strlen(val_string(val_car(v)));
}
s=cst_alloc(char,n);
p=s;
for(v=l;(v!=NULL);v=val_cdr(v))
{
if(val_stringp(val_car(v)))
{
n=strlen(val_string(val_car(v)));
memcpy(p,val_string(val_car(v)),n);
p+=n;
}
}
*p='\0';
return s;
}
static int item_match(const cst_val *PATT, const cst_val *THING,
const cst_val *sets)
{
const cst_val *sss;
if (cst_streq(val_string(PATT),val_string(THING)))
return TRUE;
for (sss=sets; sss; sss=val_cdr(sss))
{
if (cst_streq(val_string(val_car(val_car(sss))),val_string(PATT)))
{ /* Its a set not a letter */
if (val_member_string(val_string(THING),val_cdr(val_car(sss))))
{
return TRUE;
}
else
{
return FALSE;
}
}
}
return FALSE;
}
cst_utterance *default_lexical_insertion(cst_utterance *u)
{
cst_item *word;
cst_relation *sylstructure,*seg,*syl;
cst_lexicon *lex;
const cst_val *lex_addenda = NULL;
const cst_val *p, *wp = NULL;
char *phone_name;
char *stress = "0";
const char *pos;
cst_val *phones;
cst_item *ssword, *sssyl, *segitem, *sylitem, *seg_in_syl;
lex = val_lexicon(feat_val(u->features,"lexicon"));
if (lex->lex_addenda)
lex_addenda = lex->lex_addenda;
syl = utt_relation_create(u,"Syllable");
sylstructure = utt_relation_create(u,"SylStructure");
seg = utt_relation_create(u,"Segment");
for (word=relation_head(utt_relation(u,"Word"));
word; word=item_next(word))
{
ssword = relation_append(sylstructure,word);
pos = ffeature_string(word,"pos");
phones = NULL;
wp = NULL;
/* printf("awb_debug word %s pos %s gpos %s\n",
item_feat_string(word,"name"),
pos,
ffeature_string(word,"gpos")); */
/* FIXME: need to make sure that textanalysis won't split
tokens with explicit pronunciation (or that it will
propagate such to words, then we can remove the path here) */
if (item_feat_present(item_parent(item_as(word, "Token")), "phones"))
phones = (cst_val *) item_feat(item_parent(item_as(word, "Token")), "phones");
else
{
wp = val_assoc_string(item_feat_string(word, "name"),lex_addenda);
if (wp)
phones = (cst_val *)val_cdr(val_cdr(wp));
else
phones = lex_lookup(lex,item_feat_string(word,"name"),pos);
}
for (sssyl=NULL,sylitem=NULL,p=phones; p; p=val_cdr(p))
{
if (sylitem == NULL)
{
sylitem = relation_append(syl,NULL);
sssyl = item_add_daughter(ssword,sylitem);
stress = "0";
}
segitem = relation_append(seg,NULL);
phone_name = cst_strdup(val_string(val_car(p)));
if (phone_name[cst_strlen(phone_name)-1] == '1')
{
stress = "1";
phone_name[cst_strlen(phone_name)-1] = '\0';
}
else if (phone_name[cst_strlen(phone_name)-1] == '0')
{
stress = "0";
phone_name[cst_strlen(phone_name)-1] = '\0';
}
item_set_string(segitem,"name",phone_name);
seg_in_syl = item_add_daughter(sssyl,segitem);
#if 0
printf("awb_debug ph %s\n",phone_name);
#endif
if ((lex->syl_boundary)(seg_in_syl,val_cdr(p)))
{
#if 0
printf("awb_debug SYL\n");
#endif
sylitem = NULL;
if (sssyl)
item_set_string(sssyl,"stress",stress);
}
cst_free(phone_name);
}
if (!item_feat_present(item_parent(item_as(word, "Token")), "phones")
&& ! wp)
delete_val(phones);
}
return u;
}
