cst_utterance *cst_spamf0(cst_utterance *utt)
{
cst_track *spamf0_track=NULL;
cst_track *param_track=NULL;
cst_item *s;
cst_cg_db *cg_db;
cg_db = val_cg_db(utt_feat_val(utt,"cg_db"));
const cst_cart *acc_tree, *phrase_tree;
float start,end,f0val;
int num_frames,f,i;
spamf0_track=new_track();
param_track=new_track();
cst_track_resize(spamf0_track,
(utt_feat_int(utt,"param_track_num_frames")),
1);
acc_tree = cg_db->spamf0_accent_tree;
phrase_tree = cg_db->spamf0_phrase_tree;
end = 0.0;
num_frames = 0;
for (s = utt_rel_head(utt,"Segment"); s; s=item_next(s))
{
start = end;
end = start + ffeature_float(s,"segment_duration");
if(!strcmp("pau",ffeature_string(s,"name")))
{
f0val=0;
}
else
{
f0val=val_float(cart_interpret(s,phrase_tree));
}
for ( ; ((num_frames * cg_db->frame_advance) <= end) && (num_frames < utt_feat_int(utt,"param_track_num_frames")); num_frames++)
{
spamf0_track->frames[num_frames][0]=f0val;
}
}
for (s=utt_rel_head(utt,"Syllable"); s; s=item_next(s))
{
f = val_int(cart_interpret(s,acc_tree));
cst_synthtilt(cg_db,
ffeature_float(s,"R:SylStructure.daughter1.R:Segment.p.end"),
cg_db->spamf0_accent_vectors[f][0],
cg_db->spamf0_accent_vectors[f][2],
ffeature_float(s,"syllable_duration"),
cg_db->spamf0_accent_vectors[f][6],
spamf0_track);
}
param_track = val_track(utt_feat_val(utt,"param_track"));
for (i=0;i<utt_feat_int(utt,"param_track_num_frames");i++)
{
param_track->frames[i][0]=spamf0_track->frames[i][0];
}
delete_track(spamf0_track);
return utt;
}
static bool parse_track_dict( demux_t *p_demux, input_item_node_t *p_input_node,
track_elem_t *p_track, xml_reader_t *p_xml_reader,
const char *psz_element,
xml_elem_hnd_t *p_handlers )
{
VLC_UNUSED(psz_element); VLC_UNUSED(p_handlers);
input_item_t *p_new_input = NULL;
int i_ret;
p_track = new_track();
xml_elem_hnd_t track_elements[] =
// sunqueen modify start
// { {"array", COMPLEX_CONTENT, {.cmplx = skip_element} },
// {"key", SIMPLE_CONTENT, {.smpl = save_data} },
// {"integer", SIMPLE_CONTENT, {.smpl = save_data} },
// {"string", SIMPLE_CONTENT, {.smpl = save_data} },
// {"date", SIMPLE_CONTENT, {.smpl = save_data} },
{ {"array", COMPLEX_CONTENT, {(bool (__cdecl *)(track_elem_t *,const char *,char *))skip_element} },
{"key", SIMPLE_CONTENT, {save_data} },
{"integer", SIMPLE_CONTENT, {save_data} },
{"string", SIMPLE_CONTENT, {save_data} },
{"date", SIMPLE_CONTENT, {save_data} },
// sunqueen modify end
{"true", SIMPLE_CONTENT, {NULL} },
{"false", SIMPLE_CONTENT, {NULL} },
{NULL, UNKNOWN_CONTENT, {NULL} }
};
i_ret = parse_dict( p_demux, p_input_node, p_track,
p_xml_reader, "dict", track_elements );
msg_Dbg( p_demux, "name: %s, artist: %s, album: %s, genre: %s, trackNum: %s, location: %s",
p_track->name, p_track->artist, p_track->album, p_track->genre, p_track->trackNum, p_track->location );
if( !p_track->location )
{
msg_Err( p_demux, "Track needs Location" );
free_track( p_track );
return false;
}
msg_Info( p_demux, "Adding '%s'", p_track->location );
p_new_input = input_item_New( p_track->location, NULL );
input_item_node_AppendItem( p_input_node, p_new_input );
/* add meta info */
add_meta( p_new_input, p_track );
vlc_gc_decref( p_new_input );
p_demux->p_sys->i_ntracks++;
free_track( p_track );
return i_ret;
}
static void
_get_toc_callback(CDTOCDescriptor *track, void *user_data)
{
PyObject *tobj = new_track(track);
if(PyList_Check((PyObject*)user_data))
{
PyList_Append((PyObject*)user_data, tobj);
}
else
{
PyObject_CallMethod((PyObject*)user_data, "append", "O", tobj);
}
Py_XDECREF(tobj);
}
int main(int argc, char **argv)
{
cst_track *lpc;
cst_wave *sig, *sig2;
cst_sts *sts;
if (argc != 6)
{
fprintf(stderr,"usage: find_sts lpc_min lpc_range LPC WAVEFILE STS\n");
return 1;
}
lpc_min = atof(argv[1]);
lpc_range = atof(argv[2]);
lpc = new_track();
cst_track_load_est(lpc,argv[3]);
sig = new_wave();
if (cst_wave_load_riff(sig,argv[4]) == CST_WRONG_FORMAT)
{
fprintf(stderr,
"cannot load waveform, format unrecognized, from \"%s\"\n",
argv[4]);
exit(-1);
}
sts = find_sts(sig,lpc);
/* See if it worked */
sig2 = reconstruct_wave(sig,sts,lpc);
compare_waves(sig,sig2);
cst_wave_save_riff(sig2,"sig2.wav");
save_sts(sts,lpc,sig,argv[5]);
return 0;
}
cst_track *mlpg(const cst_track *param_track, cst_cg_db *cg_db)
{
/* Generate an (mcep) track using Maximum Likelihood Parameter Generation */
MLPGPARA param = NODATA;
cst_track *out;
int dim, dim_st;
// float like;
int i,j;
int nframes;
PStreamChol pst;
nframes = param_track->num_frames;
dim = (param_track->num_channels/2)-1;
dim_st = dim/2; /* dim2 in original code */
out = new_track();
cst_track_resize(out,nframes,dim_st+1);
param = xmlpgpara_init(dim,dim_st,nframes,nframes);
// mixture-index sequence
param->clsidxv = xlvalloc(nframes);
for (i=0; i<nframes; i++)
param->clsidxv->data[i] = i;
// initial static feature sequence
param->stm = xdmalloc(nframes,dim_st);
for (i=0; i<nframes; i++)
{
for (j=0; j<dim_st; j++)
param->stm->data[i][j] = param_track->frames[i][(j+1)*2];
}
/* Load cluster means */
for (i=0; i<nframes; i++)
for (j=0; j<dim_st; j++)
param->mean->data[i][j] = param_track->frames[i][(j+1)*2];
/* GMM parameters diagonal covariance */
InitPStreamChol(&pst, cg_db->dynwin, cg_db->dynwinsize, dim_st-1, nframes);
param->pdf = xdmalloc(nframes,dim*2);
param->cov = xdmalloc(nframes,dim);
for (i=0; i<nframes; i++)
for (j=0; j<dim; j++)
param->cov->data[i][j] =
param_track->frames[i][(j+1)*2+1] *
param_track->frames[i][(j+1)*2+1];
param->detvec = xget_detvec_diamat2inv(param->cov);
/* global variance parameters */
/* TBD get_gv_mlpgpara(param, vmfile, vvfile, dim2, msg_flag); */
get_dltmat(param->stm, &pst.dw, 1, param->dltm);
//like =
get_like_pdfseq_vit(dim, dim_st, nframes, nframes, param,
param_track->frames, XTRUE);
/* vlike = get_like_gv(dim2, dnum, param); */
mlgparaChol(param->pdf, &pst, param->stm);
/* Put the answer back into the output track */
for (i=0; i<nframes; i++)
{
out->times[i] = param_track->times[i];
out->frames[i][0] = param_track->frames[i][0]; /* F0 */
for (j=0; j<dim_st; j++)
out->frames[i][j+1] = param->stm->data[i][j];
}
// memory free
xmlpgparafree(param);
pst_free(&pst);
return out;
}
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;
}
int main(int argc, char **argv)
{
cst_track *t1;
cst_track *me_filters = NULL;
cst_wave *w1, *w2, *res = NULL;
cst_val *files;
cst_features *args;
int i, j;
int order, o, s;
int frame_length;
float *lpcs, *residual;
float m;
const char *f1, *f2;
const char *resfn = NULL;
int last_peak = 0, next_peak;
int period;
float power;
int rfc = 0;
int str = 0;
int fn, fo, ss;
float xpulse, xnoise;
float fxpulse, fxnoise;
float x, me;
float *hpulse = NULL, *hnoise = NULL;
float *xpulsesig = NULL, *xnoisesig = NULL;
int q = 0;
int position;
int lpc_start = 0;
args = new_features();
files =
cst_args(argv, argc,
"usage: lpc_resynth OPTIONS INTRACK OUTWAVE\n"
"Resynth an lpc track\n"
"-res <string> residual (as waveform)\n"
"-save_res Save the generated residual\n"
"-lpc_start <int> start of lpc params in lpc track {1}\n"
"-order <int> LPC order {16}\n"
"-str mixed excitation strengths\n"
"-me_filters <string> mixed excitation filters\n"
"-rfc Coefficents are reflection coefficients\n", args);
f1 = val_string(val_car(files));
f2 = val_string(val_car(val_cdr(files)));
t1 = new_track();
lpc_start = mimic_get_param_int(args, "-lpc_start", 1);
if (feat_present(args, "-rfc"))
rfc = 1;
if (feat_present(args, "-str"))
str = 1;
if (feat_present(args, "-me_filters"))
{
me_filters = new_track();
if (cst_track_load_est
(me_filters,
mimic_get_param_string(args, "-me_filters",
"me_filters.track")) != CST_OK_FORMAT)
{
fprintf(stderr,
"lpc_resynth: can't read file or wrong format \"%s\"\n",
f1);
return -1;
}
hpulse = cst_alloc(float, me_filters->num_channels);
hnoise = cst_alloc(float, me_filters->num_channels);
xpulsesig = cst_alloc(float, me_filters->num_channels);
xnoisesig = cst_alloc(float, me_filters->num_channels);
}