/* 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;
}
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;
}
int relation_load(cst_relation *r, const char *filename)
{
cst_tokenstream *fd;
cst_item *item;
const char *token=0;
if ((fd = ts_open(filename,NULL,";","","")) == 0)
{
cst_errmsg("relation_load: can't open file \"%s\" for reading\n",
filename);
return CST_ERROR_FORMAT;
}
for ( ; !ts_eof(fd); )
{
token = ts_get(fd);
if (cst_streq("#",token))
break;
}
#import "OpenEarsStaticAnalysisToggle.h"
#ifdef STATICANALYZEDEPENDENCIES
#define __clang_analyzer__ 1
#endif
#if !defined(__clang_analyzer__) || defined(STATICANALYZEDEPENDENCIES)
#undef __clang_analyzer__
if (!cst_streq("#",token))
#endif
{
cst_errmsg("relation_load: no end of header marker in \"%s\"\n",
filename);
ts_close(fd);
return CST_ERROR_FORMAT;
}
while (!ts_eof(fd))
{
token = ts_get(fd);
if (cst_streq(token,""))
continue;
item = relation_append(r,NULL);
item_set_float(item,"end",(float)cst_atof(token));
#import "OpenEarsStaticAnalysisToggle.h"
#ifdef STATICANALYZEDEPENDENCIES
#define __clang_analyzer__ 1
#endif
#if !defined(__clang_analyzer__) || defined(STATICANALYZEDEPENDENCIES)
#undef __clang_analyzer__
token = ts_get(fd);
#endif
token = ts_get(fd);
item_set_string(item,"name",token);
}
ts_close(fd);
return CST_OK_FORMAT;
}
int relation_load(cst_relation *r, const char *filename)
{
cst_tokenstream *fd;
cst_item *item;
const char *token=0;
if ((fd = ts_open(filename,NULL,";","","")) == 0)
{
cst_errmsg("relation_load: can't open file \"%s\" for reading\n",
filename);
return CST_ERROR_FORMAT;
}
for ( ; !ts_eof(fd); )
{
token = ts_get(fd);
if (cst_streq("#",token))
break;
}
if (!cst_streq("#",token))
{
cst_errmsg("relation_load: no end of header marker in \"%s\"\n",
filename);
ts_close(fd);
return CST_ERROR_FORMAT;
}
while (!ts_eof(fd))
{
token = ts_get(fd);
if (cst_streq(token,""))
continue;
item = relation_append(r,NULL);
item_set_float(item,"end",(float)cst_atof(token));
token = ts_get(fd);
token = ts_get(fd);
item_set_string(item,"name",token);
}
ts_close(fd);
return CST_OK_FORMAT;
}
static cst_utterance *cg_make_params(cst_utterance *utt)
{
/* puts in the frame items */
/* historically called "mcep" but can actually be any random vectors */
cst_cg_db *cg_db;
cst_relation *mcep, *mcep_link;
cst_item *s, *mcep_parent, *mcep_frame;
int num_frames;
float start, end;
float dur_stretch, tok_stretch;
cg_db = val_cg_db(utt_feat_val(utt,"cg_db"));
mcep = utt_relation_create(utt,"mcep");
mcep_link = utt_relation_create(utt,"mcep_link");
end = 0.0;
num_frames = 0;
dur_stretch = get_param_float(utt->features,"duration_stretch", 1.0);
for (s = utt_rel_head(utt,"HMMstate"); s; s=item_next(s))
{
start = end;
tok_stretch = ffeature_float(s,"R:segstate.parent.R:SylStructure.parent.parent.R:Token.parent.local_duration_stretch");
if (tok_stretch == 0)
tok_stretch = 1.0;
end = start + (tok_stretch*dur_stretch*cg_state_duration(s,cg_db));
item_set_float(s,"end",end);
mcep_parent = relation_append(mcep_link, s);
for ( ; (num_frames * cg_db->frame_advance) <= end; num_frames++ )
{
mcep_frame = relation_append(mcep,NULL);
item_add_daughter(mcep_parent,mcep_frame);
item_set_int(mcep_frame,"frame_number",num_frames);
item_set(mcep_frame,"name",item_feat(mcep_parent,"name"));
}
}
/* Copy duration up onto Segment relation */
for (s = utt_rel_head(utt,"Segment"); s; s=item_next(s))
item_set(s,"end",ffeature(s,"R:segstate.daughtern.end"));
utt_set_feat_int(utt,"param_track_num_frames",num_frames);
return utt;
}
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;
}
static cst_utterance *cg_predict_params(cst_utterance *utt)
{
cst_cg_db *cg_db;
cst_track *param_track;
cst_track *str_track = NULL;
cst_item *mcep;
const cst_cart *mcep_tree, *f0_tree;
int i,j,f,p,fd,o;
const char *mname;
float f0_val;
int fff;
int extra_feats = 0;
cg_db = val_cg_db(utt_feat_val(utt,"cg_db"));
param_track = new_track();
if (cg_db->do_mlpg) /* which should be the default */
fff = 1; /* copy details with stddevs */
else
fff = 2; /* copy details without stddevs */
extra_feats = 1; /* voicing */
if (cg_db->mixed_excitation)
{
extra_feats += 5;
str_track = new_track();
cst_track_resize(str_track,
utt_feat_int(utt,"param_track_num_frames"),
5);
}
cst_track_resize(param_track,
utt_feat_int(utt,"param_track_num_frames"),
(cg_db->num_channels0/fff)-
(2 * extra_feats));/* no voicing or str */
for (i=0,mcep=utt_rel_head(utt,"mcep"); mcep; i++,mcep=item_next(mcep))
{
mname = item_feat_string(mcep,"name");
for (p=0; cg_db->types[p]; p++)
if (cst_streq(mname,cg_db->types[p]))
break;
if (cg_db->types[0] == NULL)
p=0; /* if there isn't a matching tree, use the first one */
/* Predict F0 */
f0_tree = cg_db->f0_trees[p];
f0_val = val_float(cart_interpret(mcep,f0_tree));
param_track->frames[i][0] = f0_val;
/* what about stddev ? */
if (cg_db->multimodel)
{ /* MULTI model */
f = val_int(cart_interpret(mcep,cg_db->param_trees0[p]));
fd = val_int(cart_interpret(mcep,cg_db->param_trees1[p]));
item_set_int(mcep,"clustergen_param_frame",f);
param_track->frames[i][0] =
(param_track->frames[i][0]+
CG_MODEL_VECTOR(cg_db,model_vectors0,f,0)+
CG_MODEL_VECTOR(cg_db,model_vectors1,fd,0))/3.0;
for (j=2; j<param_track->num_channels; j++)
param_track->frames[i][j] =
(CG_MODEL_VECTOR(cg_db,model_vectors0,f,(j)*fff)+
CG_MODEL_VECTOR(cg_db,model_vectors1,fd,(j)*fff))/2.0;
if (cg_db->mixed_excitation)
{
o = j;
for (j=0; j<5; j++)
{
str_track->frames[i][j] =
(CG_MODEL_VECTOR(cg_db,model_vectors0,f,(o+(2*j))*fff)+
CG_MODEL_VECTOR(cg_db,model_vectors1,fd,(o+(2*j))*fff))/2.0;
}
}
}
else
{ /* SINGLE model */
/* Predict Spectral */
mcep_tree = cg_db->param_trees0[p];
f = val_int(cart_interpret(mcep,mcep_tree));
item_set_int(mcep,"clustergen_param_frame",f);
param_track->frames[i][0] =
(param_track->frames[i][0]+
CG_MODEL_VECTOR(cg_db,model_vectors0,f,0))/2.0;
for (j=2; j<param_track->num_channels; j++)
param_track->frames[i][j] =
CG_MODEL_VECTOR(cg_db,model_vectors0,f,(j)*fff);
if (cg_db->mixed_excitation)
{
o = j;
for (j=0; j<5; j++)
{
str_track->frames[i][j] =
CG_MODEL_VECTOR(cg_db,model_vectors0,f,(o+(2*j))*fff);
}
}
}
/* last coefficient is average voicing for cluster */
item_set_float(mcep,"voicing",
CG_MODEL_VECTOR(cg_db,model_vectors0,f,
cg_db->num_channels0-2));
param_track->times[i] = i * cg_db->frame_advance;
}
cg_smooth_F0(utt,cg_db,param_track);
utt_set_feat(utt,"param_track",track_val(param_track));
if (cg_db->mixed_excitation)
utt_set_feat(utt,"str_track",track_val(str_track));
return utt;
}
