/* Flite_HTS_Engine_synthesize: synthesize speech */
HTS_Boolean Flite_HTS_Engine_synthesize(Flite_HTS_Engine * f, const char *txt, const char *wav)
{
int i;
FILE *fp;
cst_voice *v = NULL;
cst_utterance *u = NULL;
cst_item *s = NULL;
char **label_data = NULL;
int label_size = 0;
if (txt == NULL)
return FALSE;
/* text analysis part */
v = REGISTER_VOX(NULL);
if (v == NULL)
return FALSE;
u = flite_synth_text(txt, v);
if (u == NULL) {
UNREGISTER_VOX(v);
return FALSE;
}
for (s = relation_head(utt_relation(u, "Segment")); s; s = item_next(s))
label_size++;
if (label_size <= 0) {
delete_utterance(u);
UNREGISTER_VOX(v);
return FALSE;
}
label_data = (char **) calloc(label_size, sizeof(char *));
for (i = 0, s = relation_head(utt_relation(u, "Segment")); s; s = item_next(s), i++) {
label_data[i] = (char *) calloc(MAXBUFLEN, sizeof(char));
Flite_HTS_Engine_create_label(f, s, label_data[i]);
}
/* speech synthesis part */
HTS_Engine_synthesize_from_strings(&f->engine, label_data, label_size);
if (wav != NULL) {
fp = fopen(wav, "wb");
HTS_Engine_save_riff(&f->engine, fp);
fclose(fp);
}
HTS_Engine_refresh(&f->engine);
for (i = 0; i < label_size; i++)
free(label_data[i]);
free(label_data);
delete_utterance(u);
UNREGISTER_VOX(v);
return TRUE;
}
/* Flite_Text_Analyzer_analysis: text analysis */
void Flite_Text_Analyzer_analysis(Flite_Text_Analyzer * analyzer, const char *text)
{
int i;
cst_item *s;
Flite_Utterance *fu;
if (analyzer == NULL || text == NULL)
return;
if (analyzer->pointer != NULL)
Flite_Text_Analyzer_clear(analyzer);
/* allocate */
fu = (Flite_Utterance *) malloc(sizeof(Flite_Utterance));
/* create voice */
fu->v = REGISTER_VOX(NULL);
if (fu->v == NULL) {
free(fu);
return;
}
/* create utterance */
fu->u = flite_synth_text(text, fu->v);
if (fu->u == NULL) {
UNREGISTER_VOX(fu->v);
free(fu);
return;
}
/* count number of phonemes */
for (fu->nitem = 0, s = relation_head(utt_relation(fu->u, "Segment")); s; s = item_next(s), fu->nitem++);
if (fu->nitem == 0) {
delete_utterance(fu->u);
UNREGISTER_VOX(fu->v);
free(fu);
return;
}
/* save informations */
fu->items = (cst_item **) malloc(sizeof(cst_item *) * fu->nitem);
for (i = 0, s = relation_head(utt_relation(fu->u, "Segment")); s; s = item_next(s), i++)
fu->items[i] = s;
analyzer->pointer = (void *) fu;
}
/* Flite_HTS_Engine_synthesis: speech synthesis */
void Flite_HTS_Engine_synthesis(Flite_HTS_Engine * f, char *txt, FILE * wavfp)
{
int i;
cst_voice *v = NULL;
cst_utterance *u = NULL;
cst_item *s = NULL;
char **label_data = NULL;
int label_size = 0;
/* text analysis part */
v = REGISTER_VOX(NULL);
if (v == NULL)
return;
u = flite_synth_text(txt, v);
if (u == NULL)
return;
for (s = relation_head(utt_relation(u, "Segment")); s; s = item_next(s))
label_size++;
if (label_size <= 0)
return;
label_data = (char **) calloc(label_size, sizeof(char *));
for (i = 0, s = relation_head(utt_relation(u, "Segment")); s;
s = item_next(s), i++) {
label_data[i] = (char *) calloc(MAXBUFLEN, sizeof(char));
Flite_HTS_Engine_create_label(f, s, label_data[i]);
}
/* speech synthesis part */
HTS_Engine_load_label_from_string_list(&f->engine, label_data, label_size);
HTS_Engine_create_sstream(&f->engine);
HTS_Engine_create_pstream(&f->engine);
HTS_Engine_create_gstream(&f->engine);
if (wavfp != NULL)
HTS_Engine_save_riff(&f->engine, wavfp);
HTS_Engine_refresh(&f->engine);
for (i = 0; i < label_size; i++)
free(label_data[i]);
free(label_data);
delete_utterance(u);
UNREGISTER_VOX(v);
}
int main(int argc, char **argv)
{
struct timeval tv;
cst_voice *v;
const char *filename;
const char *outtype;
int i;
float durs;
double time_start, time_end;
int flite_verbose, flite_loop, flite_bench;
int explicit_filename, explicit_text, explicit_phones;
#define ITER_MAX 3
int bench_iter = 0;
cst_features *extra_feats;
filename = 0;
outtype = "play"; /* default is to play */
flite_verbose = FALSE;
flite_loop = FALSE;
flite_bench = FALSE;
explicit_text = explicit_filename = explicit_phones = FALSE;
extra_feats = new_features();
flite_init();
for (i=1; i<argc; i++)
{
if (cst_streq(argv[i],"--version"))
{
flite_version();
return 1;
}
else if (cst_streq(argv[i],"-h") ||
cst_streq(argv[i],"--help") ||
cst_streq(argv[i],"-?"))
flite_usage();
else if (cst_streq(argv[i],"-v"))
flite_verbose = TRUE;
else if (cst_streq(argv[i],"-l"))
flite_loop = TRUE;
else if (cst_streq(argv[i],"-b"))
{
flite_bench = TRUE;
break; /* ignore other arguments */
}
else if ((cst_streq(argv[i],"-o")) && (i+1 < argc))
{
outtype = argv[i+1];
i++;
}
else if (cst_streq(argv[i],"-f") && (i+1 < argc))
{
filename = argv[i+1];
explicit_filename = TRUE;
i++;
}
else if (cst_streq(argv[i],"-pw"))
{
feat_set_string(extra_feats,"print_info_relation","Word");
feat_set(extra_feats,"post_synth_hook_func",
uttfunc_val(&print_info));
}
else if (cst_streq(argv[i],"-ps"))
{
feat_set_string(extra_feats,"print_info_relation","Segment");
feat_set(extra_feats,"post_synth_hook_func",
uttfunc_val(&print_info));
}
else if (cst_streq(argv[i],"-pr") && (i+1 < argc))
{
feat_set_string(extra_feats,"print_info_relation",argv[i+1]);
feat_set(extra_feats,"post_synth_hook_func",
uttfunc_val(&print_info));
i++;
}
else if ((cst_streq(argv[i],"-set") || cst_streq(argv[i],"-s"))
&& (i+1 < argc))
{
ef_set(extra_feats,argv[i+1],0);
i++;
}
else if (cst_streq(argv[i],"--seti") && (i+1 < argc))
{
ef_set(extra_feats,argv[i+1],"int");
i++;
}
else if (cst_streq(argv[i],"--setf") && (i+1 < argc))
{
ef_set(extra_feats,argv[i+1],"float");
i++;
}
else if (cst_streq(argv[i],"--sets") && (i+1 < argc))
{
ef_set(extra_feats,argv[i+1],"string");
i++;
}
else if (cst_streq(argv[i],"-p") && (i+1 < argc))
{
filename = argv[i+1];
explicit_phones = TRUE;
i++;
}
else if (cst_streq(argv[i],"-t") && (i+1 < argc))
{
filename = argv[i+1];
explicit_text = TRUE;
i++;
}
else if (filename)
outtype = argv[i];
else
filename = argv[i];
}
if (filename == NULL) filename = "-"; /* stdin */
v = REGISTER_VOX(NULL);
feat_copy_into(extra_feats,v->features);
durs = 0.0;
if (flite_bench)
{
outtype = "none";
filename = "A whole joy was reaping, but they've gone south, you should fetch azure mike.";
explicit_text = TRUE;
}
loop:
gettimeofday(&tv,NULL);
time_start = (double)(tv.tv_sec)+(((double)tv.tv_usec)/1000000.0);
if (explicit_phones)
durs = flite_phones_to_speech(filename,v,outtype);
else if ((strchr(filename,' ') && !explicit_filename) || explicit_text)
durs = flite_text_to_speech(filename,v,outtype);
else
durs = flite_file_to_speech(filename,v,outtype);
gettimeofday(&tv,NULL);
time_end = ((double)(tv.tv_sec))+((double)tv.tv_usec/1000000.0);
if (flite_verbose || (flite_bench && bench_iter == ITER_MAX))
printf("times faster than real-time: %f\n(%f seconds of speech synthesized in %f)\n",
durs/(float)(time_end-time_start),
durs,
(float)(time_end-time_start));
if (flite_loop || (flite_bench && bench_iter++ < ITER_MAX))
goto loop;
delete_features(extra_feats);
UNREGISTER_VOX(v);
return 0;
}
