int flite_voice_add_lex_addenda(cst_voice *v, const cst_string *lexfile)
{
/* Add addenda in lexfile to current voice */
cst_lexicon *lex;
const cst_val *lex_addenda = NULL;
cst_val *new_addenda;
lex = val_lexicon(feat_val(v->features,"lexicon"));
if (feat_present(v->features, "lex_addenda"))
lex_addenda = feat_val(v->features, "lex_addenda");
new_addenda = cst_lex_load_addenda(lex,lexfile);
#if 0
printf("\naddenda: ");
val_print(stdout,new_addenda);
printf("\n");
#endif
new_addenda = val_append(new_addenda,(cst_val *)lex_addenda);
if (lex->lex_addenda)
delete_val(lex->lex_addenda);
lex->lex_addenda = new_addenda;
return 0;
}
cst_utterance *cart_intonation(cst_utterance *u)
{
cst_cart *accents, *tones;
cst_item *s;
const cst_val *v;
if (feat_present(u->features,"no_intonation_accent_model"))
return u; /* not all languages have intonation models */
accents = val_cart(feat_val(u->features,"int_cart_accents"));
tones = val_cart(feat_val(u->features,"int_cart_tones"));
for (s=relation_head(utt_relation(u,"Syllable")); s; s=item_next(s))
{
v = cart_interpret(s,accents);
if (!cst_streq("NONE",val_string(v)))
item_set_string(s,"accent",val_string(v));
v = cart_interpret(s,tones);
if (!cst_streq("NONE",val_string(v)))
item_set_string(s,"endtone",val_string(v));
DPRINTF(0,("word %s gpos %s stress %s ssyl_in %s ssyl_out %s accent %s endtone %s\n",
ffeature_string(s,"R:SylStructure.parent.name"),
ffeature_string(s,"R:SylStructure.parent.gpos"),
ffeature_string(s,"stress"),
ffeature_string(s,"ssyl_in"),
ffeature_string(s,"ssyl_out"),
ffeature_string(s,"accent"),
ffeature_string(s,"endtone")));
}
return u;
}
cst_utterance *default_phrasing(cst_utterance *u)
{
cst_relation *r;
cst_item *w, *p, *lp=NULL;
const cst_val *v;
cst_cart *phrasing_cart;
r = utt_relation_create(u,"Phrase");
phrasing_cart = val_cart(feat_val(u->features,"phrasing_cart"));
for (p=NULL,w=relation_head(utt_relation(u,"Word")); w; w=item_next(w))
{
if (p == NULL)
{
p = relation_append(r,NULL);
lp = p;
#ifdef FLITE_PLUS_HTS_ENGINE
item_set_string(p,"name","BB");
#else
item_set_string(p,"name","B");
#endif /* FLITE_PLUS_HTS_ENGINE */
}
item_add_daughter(p,w);
v = cart_interpret(w,phrasing_cart);
if (cst_streq(val_string(v),"BB"))
p = NULL;
}
if (lp && item_prev(lp)) /* follow festival */
item_set_string(lp,"name","BB");
return u;
}
cst_utterance *default_pause_insertion(cst_utterance *u)
{
/* Add initial silences and silence at each phrase break */
const char *silence;
const cst_item *w;
cst_item *p, *s;
silence = val_string(feat_val(u->features,"silence"));
/* Insert initial silence */
s = relation_head(utt_relation(u,"Segment"));
if (s == NULL)
s = relation_append(utt_relation(u,"Segment"),NULL);
else
s = item_prepend(s,NULL);
item_set_string(s,"name",silence);
for (p=relation_head(utt_relation(u,"Phrase")); p; p=item_next(p))
{
for (w = item_last_daughter(p); w; w=item_prev(w))
{
s = path_to_item(w,"R:SylStructure.daughtern.daughtern.R:Segment");
if (s)
{
s = item_append(s,NULL);
item_set_string(s,"name",silence);
break;
}
}
}
return u;
}
void item_contents_set(cst_item *current, cst_item *i)
{
cst_item_contents *c = 0;
cst_item *nn_item;
if (i == 0)
c = new_item_contents(current);
else
c = i->contents;
if (c != current->contents)
{
item_unref_contents(current);
current->contents = c;
/* If this contents is already in this relation */
/* empty the other reference */
if (feat_present(current->contents->relations,current->relation->name))
{ /* oops this is already in this relation */
nn_item = val_item(feat_val(current->contents->relations,
current->relation->name));
feat_set(nn_item->contents->relations,
current->relation->name,
item_val(nn_item));
}
/* Add back reference */
feat_set(current->contents->relations,
current->relation->name,
item_val(current));
}
}
cst_utterance *default_phrasing(cst_utterance *u)
{
cst_relation *r;
cst_item *w, *p;
const cst_val *v;
cst_cart *phrasing_cart;
r = utt_relation_create(u,"Phrase");
phrasing_cart = val_cart(feat_val(u->features,"phrasing_cart"));
for (p=NULL,w=relation_head(utt_relation(u,"Word")); w; w=item_next(w))
{
if (p == NULL)
{
p = relation_append(r,NULL);
item_set_string(p,"name","BB");
}
item_add_daughter(p,w);
v = cart_interpret(w,phrasing_cart);
if (cst_streq(val_string(v),"BB"))
p = NULL;
}
return u;
}
static cst_val *cmu_LANGNAME_tokentowords(cst_item *token, const char *name)
{
/* Return list of words that expand token/name */
cst_val *r;
/* printf("token_name %s name %s\n",item_name(token),name); */
if (item_feat_present(token,"phones"))
return cons_val(string_val(name),NULL);
#if 0
if (item_feat_present(token,"nsw"))
nsw = item_feat_string(token,"nsw");
utt = item_utt(token);
lex = val_lexicon(feat_val(utt->features,"lexicon"));
#endif
if (cst_strlen(name) > 0)
r = cons_val(string_val(name),0);
else
r = NULL;
return r;
}
const cst_val *val_string_x(const char *n)
{
const cst_val *v;
/* *BUG* This will have to be fixed soon */
if (val_string_consts == NULL)
val_string_consts = new_features();
v = feat_val(val_string_consts,n);
if (v)
return v;
else
{
feat_set_string(val_string_consts,n,n);
return feat_val(val_string_consts,n);
}
}
const cst_val *get_param_val(const cst_features *f, const char *name, cst_val *def)
{
const cst_val *v;
v = feat_val(f,name);
if (v != NULL)
return v;
else
return def;
}
const char *get_param_string(const cst_features *f, const char *name, const char *def)
{
const cst_val *v;
v = feat_val(f,name);
if (v != NULL)
return val_string(v);
else
return def;
}
float get_param_float(const cst_features *f, const char *name, float def)
{
const cst_val *v;
v = feat_val(f,name);
if (v != NULL)
return val_float(v);
else
return def;
}
cst_utterance *cart_duration(cst_utterance *u)
{
cst_cart *dur_tree;
cst_item *s;
float zdur, dur_stretch, local_dur_stretch, dur;
float end;
dur_stats *ds;
const dur_stat *dur_stat;
end = 0;
if (feat_present(u->features,"no_segment_duration_model"))
return u; /* not all methods need segment durations */
dur_tree = val_cart(feat_val(u->features,"dur_cart"));
dur_stretch = get_param_float(u->features,"duration_stretch", 1.0);
ds = val_dur_stats(feat_val(u->features,"dur_stats"));
for (s=relation_head(utt_relation(u,"Segment")); s; s=item_next(s))
{
zdur = val_float(cart_interpret(s,dur_tree));
dur_stat = phone_dur_stat(ds,item_name(s));
local_dur_stretch = ffeature_float(s, "R:SylStructure.parent.parent."
"R:Token.parent.local_duration_stretch");
if (local_dur_stretch)
local_dur_stretch *= dur_stretch;
else
local_dur_stretch = dur_stretch;
dur = local_dur_stretch * ((zdur*dur_stat->stddev)+dur_stat->mean);
DPRINTF(0,("phone %s accent %s stress %s pdur %f stretch %f mean %f std %f dur %f\n",
item_name(s),
ffeature_string(s,"R:SylStructure.parent.accented"),
ffeature_string(s,"R:SylStructure.parent.stress"),
zdur, local_dur_stretch, dur_stat->mean,
dur_stat->stddev, dur));
end += dur;
item_set_float(s,"end",end);
}
return u;
}
cst_utterance *apply_synth_module(cst_utterance *u,
const cst_synth_module *mod)
{
const cst_val *v;
v = feat_val(u->features, mod->hookname);
if (v)
return (*val_uttfunc(v))(u);
if (mod->defhook)
return (*mod->defhook)(u);
return u;
}
cst_item *item_as(const cst_item *i,const char *rname)
{
/* return i as relation rname or null */
const cst_val *v;
if (i == NULL)
return NULL;
else
{
v = feat_val(i->contents->relations,rname);
if (v != NULL)
return val_item(v);
else
return NULL;
}
}
const cst_val *feat_val(const cst_features *f, const char *name)
{
cst_featvalpair *n;
n = feat_find_featpair(f, name);
if (n == NULL)
{
if (f && f->linked)
{ /* Search the linked features too if there are any */
/* We assume the linked features haven't been deleted, and */
return feat_val(f->linked, name);
}
else
return NULL; /* its really not there at all */
}
else
return n->val;
}
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;
}
float flite_file_to_speech(const char *filename,
cst_voice *voice,
const char *outtype)
{
cst_utterance *utt;
cst_tokenstream *ts;
const char *token;
cst_item *t;
cst_relation *tokrel;
float durs = 0;
int num_tokens;
cst_wave *w;
cst_breakfunc breakfunc = default_utt_break;
cst_uttfunc utt_user_callback = 0;
int fp;
if ((ts = ts_open(filename,
get_param_string(voice->features,"text_whitespace",NULL),
get_param_string(voice->features,"text_singlecharsymbols",NULL),
get_param_string(voice->features,"text_prepunctuation",NULL),
get_param_string(voice->features,"text_postpunctuation",NULL)))
== NULL)
{
cst_errmsg("failed to open file \"%s\" for reading\n",
filename);
return 1;
}
fp = get_param_int(voice->features,"file_start_position",0);
if (fp > 0)
ts_set_stream_pos(ts,fp);
if (feat_present(voice->features,"utt_break"))
breakfunc = val_breakfunc(feat_val(voice->features,"utt_break"));
if (feat_present(voice->features,"utt_user_callback"))
utt_user_callback = val_uttfunc(feat_val(voice->features,"utt_user_callback"));
/* If its a file to write to, create and save an empty wave file */
/* as we are going to incrementally append to it */
if (!cst_streq(outtype,"play") &&
!cst_streq(outtype,"none") &&
!cst_streq(outtype,"stream"))
{
w = new_wave();
cst_wave_resize(w,0,1);
cst_wave_set_sample_rate(w,16000);
cst_wave_save_riff(w,outtype); /* an empty wave */
delete_wave(w);
}
num_tokens = 0;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
while (!ts_eof(ts) || num_tokens > 0)
{
token = ts_get(ts);
if ((cst_strlen(token) == 0) ||
(num_tokens > 500) || /* need an upper bound */
(relation_head(tokrel) &&
breakfunc(ts,token,tokrel)))
{
/* An end of utt, so synthesize it */
if (utt_user_callback)
utt = (utt_user_callback)(utt);
if (utt)
{
utt = flite_do_synth(utt,voice,utt_synth_tokens);
durs += flite_process_output(utt,outtype,TRUE);
delete_utterance(utt); utt = NULL;
}
else
break;
if (ts_eof(ts)) break;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
num_tokens = 0;
}
num_tokens++;
t = relation_append(tokrel, NULL);
item_set_string(t,"name",token);
item_set_string(t,"whitespace",ts->whitespace);
item_set_string(t,"prepunctuation",ts->prepunctuation);
item_set_string(t,"punc",ts->postpunctuation);
/* Mark it at the beginning of the token */
item_set_int(t,"file_pos",
ts->file_pos-(1+ /* as we are already on the next char */
cst_strlen(token)+
cst_strlen(ts->prepunctuation)+
cst_strlen(ts->postpunctuation)));
item_set_int(t,"line_number",ts->line_number);
}
delete_utterance(utt);
ts_close(ts);
return durs;
}
const cst_phoneset *item_phoneset(const cst_item *p)
{
return val_phoneset(feat_val(item_utt(p)->features,"phoneset"));
}
cst_utterance *default_lexical_insertion(cst_utterance *u)
{
cst_item *word;
cst_relation *sylstructure,*seg,*syl;
cst_lexicon *lex, *ulex = NULL;
const cst_val *p;
char *phone_name;
char *stress = "0";
cst_val *phones;
cst_item *ssword, *sssyl, *segitem, *sylitem, *seg_in_syl;
lex = val_lexicon(feat_val(u->features,"lexicon"));
if (feat_present(u->features, "user_lexicon"))
ulex = val_lexicon(feat_val(u->features, "user_lexicon"));
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);
phones = NULL;
/* 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
{
if (ulex)
phones = lex_lookup(ulex,item_feat_string(word, "name"),0);
if (phones == NULL)
phones = lex_lookup(lex,item_feat_string(word,"name"),0);
}
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[strlen(phone_name)-1] == '1')
{
stress = "1";
phone_name[strlen(phone_name)-1] = '\0';
}
else if (phone_name[strlen(phone_name)-1] == '0')
{
stress = "0";
phone_name[strlen(phone_name)-1] = '\0';
}
item_set_string(segitem,"name",phone_name);
seg_in_syl = item_add_daughter(sssyl,segitem);
if ((lex->syl_boundary)(seg_in_syl,val_cdr(p)))
{
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"))
delete_val(phones);
}
return u;
}
const cst_val *item_feat(const cst_item *i,const char *name)
{
return feat_val(item_feats(i),name);
}
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;
}
static float flite_ssml_to_speech_ts(cst_tokenstream *ts,
cst_voice *voice,
const char *outtype)
{
cst_features *ssml_feats, *ssml_word_feats;
cst_features *attributes;
const char *token;
char *tag;
cst_utterance *utt;
cst_relation *tokrel;
int num_tokens;
cst_breakfunc breakfunc = default_utt_break;
cst_uttfunc utt_user_callback = 0;
float durs = 0.0;
cst_item *t;
ssml_feats = new_features();
ssml_word_feats = new_features();
set_charclasses(ts,
" \t\n\r",
ssml_singlecharsymbols_general,
get_param_string(voice->features,"text_prepunctuation",""),
get_param_string(voice->features,"text_postpunctuation","")
);
if (feat_present(voice->features,"utt_break"))
breakfunc = val_breakfunc(feat_val(voice->features,"utt_break"));
if (feat_present(voice->features,"utt_user_callback"))
utt_user_callback = val_uttfunc(feat_val(voice->features,"utt_user_callback"));
num_tokens = 0;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
while (!ts_eof(ts) || num_tokens > 0)
{
token = ts_get(ts);
if (cst_streq("<",token))
{ /* A tag */
tag = cst_upcase(ts_get(ts));
if (cst_streq("/",tag)) /* an end tag */
{
tag = cst_upcase(ts_get(ts));
attributes = ssml_get_attributes(ts);
feat_set_string(attributes,"_type","end");
}
else
attributes = ssml_get_attributes(ts);
utt = ssml_apply_tag(tag,attributes,utt,ssml_word_feats);
cst_free(tag);
}
else if (cst_streq("&",token))
{ /* an escape sequence */
/* skip to ; and insert value in rawdata */
}
else
{
if ((cst_strlen(token) == 0) ||
(num_tokens > 500) || /* need an upper bound */
(relation_head(tokrel) &&
breakfunc(ts,token,tokrel)))
{
/* An end of utt, so synthesize it */
if (utt_user_callback)
utt = (utt_user_callback)(utt);
if (utt)
{
utt = flite_do_synth(utt,voice,utt_synth_tokens);
durs += flite_process_output(utt,outtype,TRUE);
delete_utterance(utt); utt = NULL;
}
else
break;
if (ts_eof(ts)) break;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
num_tokens = 0;
}
num_tokens++;
t = relation_append(tokrel, NULL);
item_set_string(t,"name",token);
item_set_string(t,"whitespace",ts->whitespace);
item_set_string(t,"prepunctuation",ts->prepunctuation);
item_set_string(t,"punc",ts->postpunctuation);
/* Mark it at the beginning of the token */
item_set_int(t,"file_pos",
ts->file_pos-(1+ /* as we are already on the next char */
cst_strlen(token)+
cst_strlen(ts->prepunctuation)+
cst_strlen(ts->postpunctuation)));
item_set_int(t,"line_number",ts->line_number);
}
}
delete_utterance(utt);
return durs;
}
float feat_float(const cst_features *f, const char *name)
{
return val_float(feat_val(f,name));
}
const char *feat_string(const cst_features *f, const char *name)
{
return val_string(feat_val(f,name));
}
float flite_file_to_speech(const char *filename,
cst_voice *voice,
const char *outtype)
{
cst_utterance *utt;
cst_tokenstream *ts;
const char *token;
cst_item *t;
cst_relation *tokrel;
float d, durs = 0;
int num_tokens;
cst_breakfunc breakfunc = default_utt_break;
if ((ts = ts_open(filename,
get_param_string(voice->features,"text_whitespace",NULL),
get_param_string(voice->features,"text_singlecharsymbols",NULL),
get_param_string(voice->features,"text_prepunctuation",NULL),
get_param_string(voice->features,"text_postpunctuation",NULL)))
== NULL)
{
cst_errmsg("failed to open file \"%s\" for reading\n",
filename);
return 1;
}
if (feat_present(voice->features,"utt_break"))
breakfunc = val_breakfunc(feat_val(voice->features,"utt_break"));
/* If its a file to write to delete it as we're going to */
/* incrementally append to it */
if (!cst_streq(outtype,"play") && !cst_streq(outtype,"none"))
{
cst_wave *w;
w = new_wave();
cst_wave_resize(w,0,1);
cst_wave_set_sample_rate(w,16000);
cst_wave_save_riff(w,outtype); /* an empty wave */
delete_wave(w);
}
num_tokens = 0;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
while (!ts_eof(ts) || num_tokens > 0)
{
token = ts_get(ts);
if ((strlen(token) == 0) ||
(num_tokens > 500) || /* need an upper bound */
(relation_head(tokrel) &&
breakfunc(ts,token,tokrel)))
{
/* An end of utt */
d = flite_tokens_to_speech(utt,voice,outtype);
utt = NULL;
if (d < 0)
goto out;
durs += d;
if (ts_eof(ts))
goto out;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
num_tokens = 0;
}
num_tokens++;
t = relation_append(tokrel, NULL);
item_set_string(t,"name",token);
item_set_string(t,"whitespace",ts->whitespace);
item_set_string(t,"prepunctuation",ts->prepunctuation);
item_set_string(t,"punc",ts->postpunctuation);
item_set_int(t,"file_pos",ts->file_pos);
item_set_int(t,"line_number",ts->line_number);
}
out:
delete_utterance(utt);
ts_close(ts);
return durs;
}
