cst_utterance *f0_targets_to_pm(cst_utterance *utt)
{
cst_item *t;
float pos,lpos,f0,lf0,m;
double time;
int pm;
cst_sts_list *sts_list;
cst_lpcres *target_lpcres;
sts_list = val_sts_list(utt_feat_val(utt,"sts_list"));
lpos = 0;
lf0 = 120; /* hmm */
pm = 0;
time = 0;
/* First pass to count how many pms will be required */
for (t=relation_head(utt_relation(utt,"Target"));
t;
t=item_next(t), lf0 = f0, lpos = pos) /* changed by dhopkins */
{
pos = item_feat_float(t,"pos");
f0 = item_feat_float(t,"f0");
if (time == pos) continue;
m = (f0-lf0)/(pos-lpos);
for ( ; time < pos; pm++)
{
time += 1/(lf0 + ((time-lpos)*m));
}
}
target_lpcres = new_lpcres();
lpcres_resize_frames(target_lpcres,pm);
lpos = 0;
lf0 = 120;
pm = 0;
time = 0;
/* Second pass puts the values in */
for (t=relation_head(utt_relation(utt,"Target"));
t;
t=item_next(t), lf0 = f0, lpos = pos) /* changed by dhopkins */
{
pos = item_feat_float(t,"pos");
f0 = item_feat_float(t,"f0");
if (time == pos) continue;
m = (f0-lf0)/(pos-lpos);
for ( ; time < pos; pm++)
{
time += 1/(lf0 + ((time-lpos)*m));
target_lpcres->times[pm] = sts_list->sample_rate * time;
}
}
utt_set_feat(utt,"target_lpcres",lpcres_val(target_lpcres));
return utt;
}
static const cst_val *segment_duration(const cst_item *seg)
{
const cst_item *s = item_as(seg,"Segment");
if (!s)
return VAL_STRING_0;
else if (item_prev(s) == NULL)
return item_feat(s,"end");
else
/* It should be okay to construct this as it will get
dereferenced when the CART interpreter frees its feature
cache. */
return float_val(item_feat_float(s,"end")
- item_feat_float(item_prev(s),"end"));
}
/* 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;
}
int relation_save(cst_relation *r, const char *filename)
{
cst_file fd;
cst_item *item;
if (cst_streq(filename, "-"))
fd = stdout;
else
if ((fd = cst_fopen(filename, CST_OPEN_WRITE)) == 0)
{
cst_errmsg("relation_save: can't open file \"%s\" for writing\n",
filename);
return CST_ERROR_FORMAT;
}
for (item = relation_head(r); item; item = item_next(item))
{
if (item_feat_present(item, "end"))
cst_fprintf(fd, "%f ", item_feat_float(item, "end"));
else
cst_fprintf(fd, "%f ", 0.00);
if (item_feat_present(item, "name"))
cst_fprintf(fd, "%s ", item_feat_string(item, "name"));
else
cst_fprintf(fd, "%s ", "_");
cst_fprintf(fd, "\n");
}
if (fd != stdout)
cst_fclose(fd);
return CST_OK_FORMAT;
}
void test_hrg(void)
{
cst_utterance *u;
cst_relation *r;
cst_item *item = 0;
int i;
u = new_utterance();
r = utt_relation_create(u, "Segment");
for (i = 0; i < 10; i++)
{
char buff[20];
sprintf(buff, "seg_%03d", i);
if (i == 0)
item = relation_append(r, NULL);
else
item = item_append(item, NULL);
item_set_string(item, "name", buff);
item_set_float(item, "duration", i * 0.20);
}
for (i = 0, item = relation_head(utt_relation(u, "Segment"));
item; item = item_next(item), i++)
{
TEST_CHECK(item_feat_float(item, "duration") == correct_list[i]);
}
delete_utterance(u);
}
int main(int argc, char **argv)
{
cst_utterance *u;
cst_relation *r;
cst_item *item=0;
int i;
u = new_utterance();
r = utt_relation_create(u,"Segment");
for (i=0; i<10; i++)
{
char buff[20];
sprintf(buff,"seg_%03d",i);
if (i==0)
item = relation_append(r,NULL);
else
item = item_append(item,NULL);
item_set_string(item,"name",buff);
item_set_float(item,"duration",i*0.20);
}
for (i=0,item=relation_head(utt_relation(u,"Segment"));
item; item=item_next(item),i++)
{
printf("Segment %d %s %f\n",
i,
item_feat_string(item,"name"),
item_feat_float(item,"duration"));
}
delete_utterance(u);
return 0;
}
static int voiced_frame(cst_item *m)
{
const char *ph_vc;
const char *ph_name;
ph_vc = ffeature_string(m,"R:mcep_link.parent.R:segstate.parent.ph_vc");
ph_name = ffeature_string(m,"R:mcep_link.parent.R:segstate.parent.name");
if (cst_streq(ph_name,"pau"))
return 0; /* unvoiced */
else if (cst_streq("+",ph_vc))
return 1; /* voiced */
else if (item_feat_float(m,"voicing") > 0.5)
/* Even though the range is 0-10, I *do* mean 0.5 */
return 1; /* voiced */
else
return 0; /* unvoiced */
}
string getPhonemes( const char* sText )
{
string sRet;
cst_features* args = new_features();
cst_voice* v;
cst_utterance* u;
cst_item* s;
const char* name;
//const cst_val* d;
flite_init();
v = register_cmu_us_no_wave( NULL );
u = flite_synth_text( sText, v );
for ( s = relation_head( utt_relation( u, "Segment" ) ); s; s = item_next( s ) )
{
sRet += item_feat_string( s, "name" );
float test = item_feat_float( s, "end" );
//d = segment_duration( s );
/* If its a vowel and is stressed output stress value */
if ( ( cst_streq( "+", ffeature_string( s, "ph_vc" ) ) ) &&
( cst_streq( "1", ffeature_string( s, "R:SylStructure.parent.stress" ) ) ) )
{
sRet += "1";
}
sRet += " ";
}
delete_utterance( u );
delete_features( args );
return sRet;
}
float flite_text_to_speech_phenome( const char* text,
cst_voice* voice,
const char* outtype,
void* pStream )
{
cst_utterance* u;
float dur;
float end_last = 0;
float end_current = 0;
float dur_current = 0;
float dur_sum = 0;
//feat_set_float( voice->features, "duration_stretch", 1 );
u = flite_synth_text( text, voice );
cst_item* s;
string sRet;
int nPhoneme = 0;
for ( s = relation_head( utt_relation( u, "Segment" ) ); s; s = item_next( s ) )
{
SPhenomeTiming ps;
string sPhoneme = item_feat_string( s, "name" );
sRet += sPhoneme;
end_current = item_feat_float( s, "end" );
dur_current = end_current - end_last;
//if ( !( nPhoneme == 0 && sPhoneme == "pau" ) )
//{
dur_sum += dur_current;
//}
ps.fWeight = 1;
/* If its a vowel and is stressed output stress value */
if ( ( cst_streq( "+", ffeature_string( s, "ph_vc" ) ) ) &&
( cst_streq( "1", ffeature_string( s, "R:SylStructure.parent.stress" ) ) ) )
{
sRet += "1";
ps.fWeight = 1.3;
}
sRet += " ";
if ( pStream )
{
// fade into each other
ps.sName = sPhoneme;
ps.fStart = end_current - dur_current;
ps.fEnd = end_current;
ps.fDuration = dur_current;
( ( CryMT::queue<SPhenomeTiming>* )pStream )->push( ps );
}
end_last = end_current;
++nPhoneme;
}
dur = flite_process_output( u, outtype, FALSE );
delete_utterance( u );
return dur;
}
int main(int argc, char *argv[])
{
char *s,*fn;
cst_voice *voice; // synthesis voice
cst_utterance *utt; // current utterance
cst_wave *cstwave; // synthesised wave
Wave w; // HTK wave
short *p;
HTime sampPeriod = 625.0;
int n;
MemHeap mem;
AudioOut a;
try {
if (InitHTK(argc,argv,version)<SUCCESS){
ReportErrors("Main",0); exit(-1);
}
if (NumArgs() !=2) {
printf("SFliteTest synthstring file\n"); exit(0);
}
CreateHeap(&mem,"heap",MSTAK,1,0.0,10000,100000);
s = GetStrArg();
fn = GetStrArg();
printf("Synth: %s -> %s\n",s,fn);
// initialise Edinburgh cst lib
cst_regex_init();
// setup the voice
voice = register_cmu_us_kal16(NULL);
// convert text to waveform
utt = flite_synth_text(s,voice);
if (utt==NULL) {
HRError(12001,"SFliteTest: cant synthesise %s\n",s);
throw ATK_Error(12001);
}
cstwave = utt_wave(utt);
p = cstwave->samples; n = cstwave->num_samples;
w = OpenWaveOutput(&mem,&sampPeriod,n);
printf("%d samples created\n",n);
PutWaveSample(w,n,p);
if (CloseWaveOutput(w,WAV,fn)<SUCCESS){
ReportErrors("Main",0); exit(-1);
}
// explore structure
const cst_item *it, *itlast = NULL;
float x,y;
int i;
string lastword="0"; x = 0;
for (i=1,it = relation_head(utt_relation(utt, "Segment")); it!=NULL; it = item_next(it),i++)
{
printf("Segment %d\n",i);
y = item_feat_float(it,"end");
string ph = string(ffeature_string(it,"p.name"));
string wd = string(ffeature_string(it,"R:SylStructure.parent.parent.name"));
//printf("end = %f ph=%s wd=%s\n",y,ph.c_str(),wd.c_str());
if (wd != lastword){
printf("**** end of %s = %f\n",lastword.c_str(),x);
lastword=wd;
}
x = y;
}
//if (itlast!=NULL) {
// word = string(ffeature_string(itlast,"R:SylStructure.parent.parent.name"));
// idx = text.find(word);
//}
return 0;
}
catch (ATK_Error e){ ReportErrors("ATK",e.i); }
catch (HTK_Error e){ ReportErrors("HTK",e.i); }
return 0;
}