void OpenJTalk_initialize(OpenJTalk * open_jtalk, HTS_Boolean use_lpf, int sampling_rate,
int fperiod, double alpha, int stage, double beta, int audio_buff_size,
double uv_threshold, HTS_Boolean use_log_gain, double gv_weight_mgc,
double gv_weight_lf0, double gv_weight_lpf)
{
Mecab_initialize(&open_jtalk->mecab);
NJD_initialize(&open_jtalk->njd);
JPCommon_initialize(&open_jtalk->jpcommon);
if (use_lpf)
HTS_Engine_initialize(&open_jtalk->engine, 3);
else
HTS_Engine_initialize(&open_jtalk->engine, 2);
HTS_Engine_set_sampling_rate(&open_jtalk->engine, sampling_rate);
HTS_Engine_set_fperiod(&open_jtalk->engine, fperiod);
HTS_Engine_set_alpha(&open_jtalk->engine, alpha);
HTS_Engine_set_gamma(&open_jtalk->engine, stage);
HTS_Engine_set_log_gain(&open_jtalk->engine, use_log_gain);
HTS_Engine_set_beta(&open_jtalk->engine, beta);
HTS_Engine_set_audio_buff_size(&open_jtalk->engine, audio_buff_size);
HTS_Engine_set_msd_threshold(&open_jtalk->engine, 1, uv_threshold);
HTS_Engine_set_gv_weight(&open_jtalk->engine, 0, gv_weight_mgc);
HTS_Engine_set_gv_weight(&open_jtalk->engine, 1, gv_weight_lf0);
if (use_lpf)
HTS_Engine_set_gv_weight(&open_jtalk->engine, 2, gv_weight_lpf);
}
static int setup(struct app *app)
{
#ifdef HTS_MELP
#define NR_STREAMS 3
#else
#define NR_STREAMS 2
#endif /* HTS_MELP */
double gv_weight[] = {
app->gv_weight_mgc,
app->gv_weight_lf0,
#ifdef HTS_MELP
app->gv_weight_lpf
#endif /* HTS_MELP */
};
int i;
app->play_h = play_init(&app->play_info, "default",
SND_PCM_FORMAT_S16_LE, 1, app->sampling_rate,
500000, 8);
Mecab_initialize(&app->mecab);
if (Mecab_load(&app->mecab, app->dn_mecab) != TRUE)
return -1;
NJD_initialize(&app->njd);
JPCommon_initialize(&app->jpcommon);
HTS_Engine_initialize(&app->engine);
if (HTS_Engine_load(&app->engine, &app->fn_voice, 1) != TRUE)
return -1;
HTS_Engine_set_sampling_frequency(&app->engine,
(size_t)app->sampling_rate);
if (app->fperiod >= 0)
HTS_Engine_set_fperiod(&app->engine, app->fperiod);
if (app->alpha >= 0.0)
HTS_Engine_set_alpha(&app->engine, app->alpha);
if (app->beta >= 0.0)
HTS_Engine_set_beta(&app->engine, app->beta);
if (app->half_tone >= 0.0)
HTS_Engine_add_half_tone(&app->engine, app->half_tone);
if (app->audio_buff_size > 0)
HTS_Engine_set_audio_buff_size(&app->engine,
app->audio_buff_size);
if (app->uv_threshold >= 0.0)
HTS_Engine_set_msd_threshold(&app->engine, 1,
app->uv_threshold);
if (app->speed >= 0.0)
HTS_Engine_set_speed(&app->engine, app->speed);
for (i = 0; i < NR_STREAMS; i++)
if (gv_weight[i] >= 0.0)
HTS_Engine_set_gv_weight(&app->engine, i, gv_weight[i]);
return 0;
}
void TextToSpeech::initialize(
int sampling_rate, int fperiod, double alpha, int stage, double beta, int audio_buff_size,
double uv_threshold, HTS_Boolean use_log_gain, double gv_weight_mgc,
double gv_weight_lf0, double gv_weight_lpf)
{
Mecab_initialize(&open_jtalk_.mecab);
NJD_initialize(&open_jtalk_.njd);
JPCommon_initialize(&open_jtalk_.jpcommon);
HTS_Engine_initialize(&open_jtalk_.engine, 2);
HTS_Engine_set_sampling_rate(&open_jtalk_.engine, sampling_rate);
HTS_Engine_set_fperiod(&open_jtalk_.engine, fperiod);
HTS_Engine_set_alpha(&open_jtalk_.engine, alpha);
HTS_Engine_set_gamma(&open_jtalk_.engine, stage);
HTS_Engine_set_log_gain(&open_jtalk_.engine, use_log_gain);
HTS_Engine_set_beta(&open_jtalk_.engine, beta);
HTS_Engine_set_audio_buff_size(&open_jtalk_.engine, audio_buff_size);
HTS_Engine_set_msd_threshold(&open_jtalk_.engine, 1, uv_threshold);
HTS_Engine_set_gv_weight(&open_jtalk_.engine, 0, gv_weight_mgc);
HTS_Engine_set_gv_weight(&open_jtalk_.engine, 1, gv_weight_lf0);
}
/* Flite_HTS_Engine_initialize: initialize system */
void Flite_HTS_Engine_initialize(Flite_HTS_Engine * f,
int sampling_rate,
int fperiod,
double alpha,
int stage,
double beta,
int audio_buff_size,
double uv_threshold,
HTS_Boolean use_log_gain,
double gv_weight_mcp, double gv_weight_lf0)
{
HTS_Engine_initialize(&f->engine, 2);
HTS_Engine_set_sampling_rate(&f->engine, sampling_rate);
HTS_Engine_set_fperiod(&f->engine, fperiod);
HTS_Engine_set_alpha(&f->engine, alpha);
HTS_Engine_set_gamma(&f->engine, stage);
HTS_Engine_set_log_gain(&f->engine, use_log_gain);
HTS_Engine_set_beta(&f->engine, beta);
HTS_Engine_set_audio_buff_size(&f->engine, audio_buff_size);
HTS_Engine_set_msd_threshold(&f->engine, 1, uv_threshold);
HTS_Engine_set_gv_weight(&f->engine, 0, gv_weight_mcp);
HTS_Engine_set_gv_weight(&f->engine, 1, gv_weight_lf0);
printf("Initialized HTS Engine\n");
}
int OpenJTalk_Create(OpenJTalk** openjtalk,Mecab* mecab,int argc,const char **argv)
{
/* file names of models */
char *fn_ms_dur = NULL;
char *fn_ms_mgc = NULL;
char *fn_ms_lf0 = NULL;
char *fn_ms_lpf = NULL;
/* file names of trees */
char *fn_ts_dur = NULL;
char *fn_ts_mgc = NULL;
char *fn_ts_lf0 = NULL;
char *fn_ts_lpf = NULL;
/* file names of windows */
int num_ws_mgc = 0, num_ws_lf0 = 0, num_ws_lpf = 0;
/* file names of global variance */
char *fn_ms_gvm = NULL;
char *fn_ms_gvl = NULL;
char *fn_ms_gvf = NULL;
/* file names of global variance trees */
char *fn_ts_gvm = NULL;
char *fn_ts_gvl = NULL;
char *fn_ts_gvf = NULL;
/* file names of global variance switch */
char *fn_gv_switch = NULL;
/* global parameter */
int sampling_rate = 16000;
int fperiod = 80;
double alpha = 0.42;
int stage = 0; /* gamma = -1.0/stage */
double beta = 0.0;
int audio_buff_size = 1600;
double uv_threshold = 0.5;
double gv_weight_mgc = 1.0;
double gv_weight_lf0 = 1.0;
double gv_weight_lpf = 1.0;
HTS_Boolean use_log_gain = FALSE;
HTS_Boolean use_lpf = FALSE;
/* alloc OpenJTalk ythis pointer! */
*openjtalk = (struct OpenJTalk*)malloc(sizeof(struct OpenJTalk));
(*openjtalk)->mecab = mecab;
(*openjtalk)->errorout[0] = '\0';
/* delta window handler for mel-cepstrum */
(*openjtalk)->fn_ws_mgc = (char **) calloc(argc, sizeof(char *));
/* delta window handler for log f0 */
(*openjtalk)->fn_ws_lf0 = (char **) calloc(argc, sizeof(char *));
/* delta window handler for low-pass filter */
(*openjtalk)->fn_ws_lpf = (char **) calloc(argc, sizeof(char *));
/* parse command line */
if (argc == 1)
{
sprintf((*openjtalk)->errorout,"option error!");
return -1;
}
/* read command */
while (--argc) {
if (**++argv == '-') {
switch (*(*argv + 1)) {
case 't':
switch (*(*argv + 2)) {
case 'd':
fn_ts_dur = (char*)*(++argv);
break;
case 'm':
fn_ts_mgc = (char*)*(++argv);
break;
case 'f':
case 'p':
fn_ts_lf0 = (char*)*(++argv);
break;
case 'l':
fn_ts_lpf = (char*)*(++argv);
break;
default:
sprintf((*openjtalk)->errorout,"Invalid option '-t%c'.\n", *(*argv + 2));
return 0;
}
--argc;
break;
case 'm':
switch (*(*argv + 2)) {
case 'd':
fn_ms_dur = (char*)*(++argv);
break;
case 'm':
fn_ms_mgc = (char*)*(++argv);
break;
case 'f':
case 'p':
fn_ms_lf0 = (char*)*(++argv);
break;
case 'l':
fn_ms_lpf = (char*)*(++argv);
break;
default:
sprintf((*openjtalk)->errorout,"Invalid option '-m%c'.\n", *(*argv + 2));
return 0;
}
--argc;
break;
case 'd':
switch (*(*argv + 2)) {
case 'm':
(*openjtalk)->fn_ws_mgc[num_ws_mgc++] = (char*)*(++argv);
break;
case 'f':
case 'p':
(*openjtalk)->fn_ws_lf0[num_ws_lf0++] = (char*)*(++argv);
break;
case 'l':
(*openjtalk)->fn_ws_lpf[num_ws_lpf++] = (char*)*(++argv);
break;
default:
sprintf((*openjtalk)->errorout,"Invalid option '-d%c'.\n", *(*argv + 2));
return 0;
}
--argc;
break;
case 's':
sampling_rate = atoi(*++argv);
--argc;
break;
case 'p':
fperiod = atoi(*++argv);
--argc;
break;
case 'a':
alpha = atof(*++argv);
--argc;
break;
case 'g':
stage = atoi(*++argv);
--argc;
break;
case 'l':
use_log_gain = TRUE;
break;
case 'b':
beta = atof(*++argv);
--argc;
break;
case 'u':
uv_threshold = atof(*++argv);
--argc;
break;
case 'e':
switch (*(*argv + 2)) {
case 'm':
fn_ts_gvm = (char*)*(++argv);
break;
case 'f':
case 'p':
fn_ts_gvl = (char*)*(++argv);
break;
case 'l':
fn_ts_gvf = (char*)*(++argv);
break;
default:
sprintf((*openjtalk)->errorout,"Invalid option '-e%c'.\n", *(*argv + 2));
return 0;
}
--argc;
break;
case 'c':
switch (*(*argv + 2)) {
case 'm':
fn_ms_gvm = (char*)*(++argv);
break;
case 'f':
case 'p':
fn_ms_gvl = (char*)*(++argv);
break;
case 'l':
fn_ms_gvf = (char*)*(++argv);
break;
default:
sprintf((*openjtalk)->errorout,"Invalid option '-c%c'.\n", *(*argv + 2));
return 0;
}
--argc;
break;
case 'j':
switch (*(*argv + 2)) {
case 'm':
gv_weight_mgc = atof(*(++argv));
break;
case 'f':
case 'p':
gv_weight_lf0 = atof(*(++argv));
break;
case 'l':
gv_weight_lpf = atof(*(++argv));
break;
default:
sprintf((*openjtalk)->errorout,"Invalid option '-j%c'.\n", *(*argv + 2));
return 0;
}
--argc;
break;
case 'k':
fn_gv_switch = (char*)*++argv;
--argc;
break;
case 'z':
audio_buff_size = atoi(*++argv);
--argc;
break;
default:
sprintf((*openjtalk)->errorout,"Invalid option '-%c'.\n", *(*argv + 1));
return 0;
}
}
}
/* number of models,trees check */
if (fn_ms_dur == NULL || fn_ms_mgc == NULL || fn_ms_lf0 == NULL ||
fn_ts_dur == NULL || fn_ts_mgc == NULL || fn_ts_lf0 == NULL ||
num_ws_mgc <= 0 || num_ws_lf0 <= 0) {
sprintf((*openjtalk)->errorout,"Specify models (trees) for each parameter.\n");
return 0;
}
if (fn_ms_lpf != NULL && fn_ts_lpf != NULL && num_ws_lpf > 0)
use_lpf = TRUE;
/* initialize */
NJD_initialize(&(*openjtalk)->njd);
JPCommon_initialize(&(*openjtalk)->jpcommon);
if (use_lpf)
HTS_Engine_initialize(&(*openjtalk)->engine, 3);
else
HTS_Engine_initialize(&(*openjtalk)->engine, 2);
HTS_Engine_set_sampling_rate(&(*openjtalk)->engine, sampling_rate);
HTS_Engine_set_fperiod(&(*openjtalk)->engine, fperiod);
HTS_Engine_set_alpha(&(*openjtalk)->engine, alpha);
HTS_Engine_set_gamma(&(*openjtalk)->engine, stage);
HTS_Engine_set_log_gain(&(*openjtalk)->engine, use_log_gain);
HTS_Engine_set_beta(&(*openjtalk)->engine, beta);
HTS_Engine_set_audio_buff_size(&(*openjtalk)->engine, audio_buff_size);
HTS_Engine_set_msd_threshold(&(*openjtalk)->engine, 1, uv_threshold);
HTS_Engine_set_gv_weight(&(*openjtalk)->engine, 0, gv_weight_mgc);
HTS_Engine_set_gv_weight(&(*openjtalk)->engine, 1, gv_weight_lf0);
if (use_lpf)
{
HTS_Engine_set_gv_weight(&(*openjtalk)->engine, 2, gv_weight_lpf);
}
/* Load! */
HTS_Engine_load_duration_from_fn(&(*openjtalk)->engine, &fn_ms_dur, &fn_ts_dur, 1);
HTS_Engine_load_parameter_from_fn(&(*openjtalk)->engine, &fn_ms_mgc, &fn_ts_mgc, (*openjtalk)->fn_ws_mgc, 0,
FALSE, num_ws_mgc, 1);
HTS_Engine_load_parameter_from_fn(&(*openjtalk)->engine, &fn_ms_lf0, &fn_ts_lf0, (*openjtalk)->fn_ws_lf0, 1,
TRUE, num_ws_lf0, 1);
if (HTS_Engine_get_nstream(&(*openjtalk)->engine) == 3)
HTS_Engine_load_parameter_from_fn(&(*openjtalk)->engine, &fn_ms_lpf, &fn_ts_lpf, (*openjtalk)->fn_ws_lpf, 2,
FALSE, num_ws_lpf, 1);
if (fn_ms_gvm != NULL) {
if (fn_ts_gvm != NULL)
HTS_Engine_load_gv_from_fn(&(*openjtalk)->engine, &fn_ms_gvm, &fn_ts_gvm, 0, 1);
else
HTS_Engine_load_gv_from_fn(&(*openjtalk)->engine, &fn_ms_gvm, NULL, 0, 1);
}
if (fn_ms_gvl != NULL) {
if (fn_ts_gvl != NULL)
HTS_Engine_load_gv_from_fn(&(*openjtalk)->engine, &fn_ms_gvl, &fn_ts_gvl, 1, 1);
else
HTS_Engine_load_gv_from_fn(&(*openjtalk)->engine, &fn_ms_gvl, NULL, 1, 1);
}
if (HTS_Engine_get_nstream(&(*openjtalk)->engine) == 3 && fn_ms_gvf != NULL) {
if (fn_ts_gvf != NULL)
HTS_Engine_load_gv_from_fn(&(*openjtalk)->engine, &fn_ms_gvf, &fn_ts_gvf, 2, 1);
else
HTS_Engine_load_gv_from_fn(&(*openjtalk)->engine, &fn_ms_gvf, NULL, 2, 1);
}
if (fn_gv_switch != NULL)
HTS_Engine_load_gv_switch_from_fn(&(*openjtalk)->engine, fn_gv_switch);
return 1;
}
int main(int argc, char* argv[]) {
size_t num_voices;
char **fn_voices;
char* in_fname;
char* output_fname;
FILE * outfp;
char* dur_fname;
FILE * durfp;
bool print_label = false;
bool print_utt = false;
bool write_raw = false;
bool write_durlabel = false;
CFSAString LexFileName, LexDFileName;
HTS_Engine engine;
double speed = 1.1;
size_t fr = 48000;
size_t fp = 240;
float alpha = 0.55;
float beta = 0.0;
float ht = 2.0;
float th = 0.5;
float gvw1 = 1.0;
float gvw2 = 1.2;
FSCInit();
fn_voices = (char **) malloc(argc * sizeof (char *));
if (argc < 11) {
fprintf(stderr, "Viga: liiga vähe parameetreid\n\n");
PrintUsage();
}
for (int i = 0; i < argc; i++) {
if (CFSAString("-lex") == argv[i]) {
if (i + 1 < argc) {
LexFileName = argv[++i];
} else {
return PrintUsage();
}
}
if (CFSAString("-lexd") == argv[i]) {
if (i + 1 < argc) {
LexDFileName = argv[++i];
} else {
return PrintUsage();
}
}
if (CFSAString("-m") == argv[i]) {
if (i + 1 < argc) {
fn_voices[0] = argv[i + 1];
} else {
fprintf(stderr, "Viga: puudub *.htsvoice fail\n");
PrintUsage();
exit(0);
}
}
if (CFSAString("-o") == argv[i]) {
if (i + 1 < argc) {
output_fname = argv[i + 1];
cfileexists(output_fname);
} else {
fprintf(stderr, "Viga: puudb väljundfaili nimi\n");
PrintUsage();
exit(0);
}
}
if (CFSAString("-f") == argv[i]) {
if (i + 1 < argc) {
in_fname = argv[i + 1];
} else {
fprintf(stderr, "Viga: puudb sisendfaili nimi\n");
PrintUsage();
exit(0);
}
}
if (CFSAString("-s") == argv[i]) {
if (i + 1 < argc) {
samplerate(fr, fp, alpha, atoi(argv[i + 1]));
}
}
if (CFSAString("-r") == argv[i]) {
if (i + 1 < argc) {
speed = atof(argv[i + 1]);
}
}
if (CFSAString("-ht") == argv[i]) {
if (i + 1 < argc) {
ht = atof(argv[i + 1]);
}
}
if (CFSAString("-gvw1") == argv[i]) {
if (i + 1 < argc) {
gvw1 = atof(argv[i + 1]);
}
}
if (CFSAString("-gvw2") == argv[i]) {
if (i + 1 < argc) {
gvw2 = atof(argv[i + 1]);
}
}
if (CFSAString("-debug") == argv[i]) {
print_label = true;
}
if (CFSAString("-utt") == argv[i]) {
print_utt = true;
}
if (CFSAString("-raw") == argv[i]) {
write_raw = true;
}
if (CFSAString("-dur") == argv[i]) {
if (i + 1 < argc) {
dur_fname = argv[i + 1];
cfileexists(dur_fname);
write_durlabel = true;
} else {
fprintf(stderr, "Viga: puudb kestustefaili nimi\n");
PrintUsage();
exit(0);
}
}
}
Linguistic.Open(LexFileName);
Disambiguator.Open(LexDFileName);
CFSWString text;
ReadUTF8Text(text, in_fname);
HTS_Engine_initialize(&engine);
if (HTS_Engine_load(&engine, fn_voices, 1) != TRUE) {
fprintf(stderr, "Viga: puudub *.htsvoice. %p\n", fn_voices[0]);
free(fn_voices);
HTS_Engine_clear(&engine);
exit(1);
}
free(fn_voices);
HTS_Engine_set_sampling_frequency(&engine, (size_t) fr);
HTS_Engine_set_phoneme_alignment_flag(&engine, FALSE);
HTS_Engine_set_fperiod(&engine, (size_t) fp);
HTS_Engine_set_alpha(&engine, alpha);
HTS_Engine_set_beta(&engine, beta);
HTS_Engine_set_speed(&engine, speed);
HTS_Engine_add_half_tone(&engine, ht);
HTS_Engine_set_msd_threshold(&engine, 1, th);
/*
HTS_Engine_set_duration_interpolation_weight(&engine, 1, diw);
HTS_Engine_set_parameter_interpolation_weight(&engine, 0, 0, piw1);
HTS_Engine_set_parameter_interpolation_weight(&engine, 0, 1, piw2);
HTS_Engine_set_gv_interpolation_weight(&engine, 0, 0, giw1);
HTS_Engine_set_gv_interpolation_weight(&engine, 0, 1, giw2);
*/
HTS_Engine_set_gv_weight(&engine, 0, gvw1);
HTS_Engine_set_gv_weight(&engine, 1, gvw2);
text = DealWithText(text);
CFSArray<CFSWString> res = do_utterances(text);
INTPTR data_size = 0;
outfp = fopen(output_fname, "wb");
if (write_durlabel) durfp = fopen(dur_fname, "w");
if (!write_raw) HTS_Engine_write_header(&engine, outfp, 1);
for (INTPTR i = 0; i < res.GetSize(); i++) {
CFSArray<CFSWString> label = do_all(res[i], print_label, print_utt);
std::vector<std::string> v;
v = to_vector(label);
std::vector<char*> vc;
fill_char_vector(v, vc);
size_t n_lines = vc.size();
if (HTS_Engine_synthesize_from_strings(&engine, &vc[0], n_lines) != TRUE) {
fprintf(stderr, "Viga: süntees ebaonnestus.\n");
HTS_Engine_clear(&engine);
exit(1);
}
clean_char_vector(vc);
data_size += HTS_Engine_engine_speech_size(&engine);
if (write_durlabel) HTS_Engine_save_durlabel(&engine, durfp);
HTS_Engine_save_generated_speech(&engine, outfp);
HTS_Engine_refresh(&engine);
} //synth loop
if (!write_raw) HTS_Engine_write_header(&engine, outfp, data_size);
if (write_durlabel) fclose(durfp);
fclose(outfp);
HTS_Engine_clear(&engine);
Linguistic.Close();
FSCTerminate();
return 0;
}
/* Flite_HTS_Engine_set_alpha: set alpha */
void Flite_HTS_Engine_set_alpha(Flite_HTS_Engine * f, double d)
{
HTS_Engine_set_alpha(&f->engine, d);
}
static void Open_JTalk_set_alpha(Open_JTalk * open_jtalk, double f)
{
HTS_Engine_set_alpha(&open_jtalk->engine, f);
}
static void Initialize(SUttProcessor *self, const SVoice *voice, s_erc *error)
{
hts_params *engine_params;
SHTSEngineMESynthUttProc105 *HTSsynth = (SHTSEngineMESynthUttProc105*)self;
const SMap *hts_data;
const SObject *vcfgObject;
char *voice_base_path;
S_CLR_ERR(error);
/* get voice base path */
vcfgObject = SVoiceGetFeature(voice, "config_file", error);
if (S_CHK_ERR(error, S_CONTERR,
"Initialize",
"Call to \"SVoiceGetFeature\" failed, failed to get voice config file"))
return;
voice_base_path = s_get_base_path(SObjectGetString(vcfgObject, error), error);
if (S_CHK_ERR(error, S_CONTERR,
"Initialize",
"Call to \"s_get_base_path/SObjectGetString\" failed"))
return;
/* get the HTS engine settings */
engine_params = get_hts_engine_params(self->features, &(HTSsynth->me), error);
if (S_CHK_ERR(error, S_CONTERR,
"Initialize",
"Call to \"get_hts_engine_params\" failed"))
{
S_FREE(voice_base_path);
return;
}
/* initialize the engine */
if (HTSsynth->me == TRUE)
{
/* extra stream for strengths */
HTS_Engine_initialize(&(HTSsynth->engine), 3);
}
else
{
HTS_Engine_initialize(&(HTSsynth->engine), 2);
}
/* set the engine parameters */
HTS_Engine_set_sampling_rate(&(HTSsynth->engine), engine_params->sampling_rate);
HTS_Engine_set_fperiod(&(HTSsynth->engine), engine_params->fperiod);
HTS_Engine_set_alpha(&(HTSsynth->engine), engine_params->alpha);
HTS_Engine_set_gamma(&(HTSsynth->engine), engine_params->stage);
HTS_Engine_set_log_gain(&(HTSsynth->engine), engine_params->use_log_gain);
HTS_Engine_set_beta(&(HTSsynth->engine), engine_params->beta);
HTS_Engine_set_audio_buff_size(&(HTSsynth->engine), engine_params->audio_buff_size);
HTS_Engine_set_msd_threshold(&(HTSsynth->engine), 1, engine_params->uv_threshold);
HTS_Engine_set_gv_weight(&(HTSsynth->engine), 0, engine_params->gv_weight_mcp);
HTS_Engine_set_gv_weight(&(HTSsynth->engine), 1, engine_params->gv_weight_lf0);
if (HTSsynth->me == TRUE)
HTS_Engine_set_gv_weight(&(HTSsynth->engine), 2, engine_params->gv_weight_str);
S_FREE(engine_params);
hts_data = S_MAP(SVoiceGetFeature(voice, "hts engine data", error));
if (S_CHK_ERR(error, S_CONTERR,
"Initialize",
"Call to \"SVoiceGetFeature\" failed"))
goto quit_error;
if (hts_data == NULL)
{
S_CTX_ERR(error, S_FAILURE,
"Initialize",
"Failed to get \"hts engine data\" map from voice features");
goto quit_error;
}
load_hts_engine_data(hts_data, HTSsynth, voice_base_path, error);
if (S_CHK_ERR(error, S_CONTERR,
"Initialize",
"Call to \"load_hts_engine_data\" failed"))
goto quit_error;
HTS_Engine_set_duration_interpolation_weight(&(HTSsynth->engine), 0, 1.0);
HTS_Engine_set_parameter_interpolation_weight(&(HTSsynth->engine), 0, 0, 1.0);
HTS_Engine_set_parameter_interpolation_weight(&(HTSsynth->engine), 1, 0, 1.0);
if (HTSsynth->me == TRUE)
HTS_Engine_set_parameter_interpolation_weight(&(HTSsynth->engine), 2, 0, 1.0);
HTS_Engine_set_gv_interpolation_weight(&(HTSsynth->engine), 0, 0, 1.0);
HTS_Engine_set_gv_interpolation_weight(&(HTSsynth->engine), 1, 0, 1.0);
if (HTSsynth->me == TRUE)
HTS_Engine_set_gv_interpolation_weight(&(HTSsynth->engine), 2, 0, 1.0);
/* all OK */
S_FREE(voice_base_path);
return;
/* error clean up */
quit_error:
HTS_Engine_clear(&(HTSsynth->engine));
filter_destructor(HTSsynth);
if (voice_base_path != NULL)
S_FREE(voice_base_path);
}
int htsSynthesize(int argc, char **argv)
{
int i;
double f;
/* hts_engine API */
HTS_Engine engine;
/* HTS voices */
size_t num_voices;
char **fn_voices;
/* input label file name */
char *labfn = NULL;
/* output file pointers */
FILE *durfp = NULL, *mgcfp = NULL, *lf0fp = NULL, *lpffp = NULL, *wavfp = NULL, *rawfp = NULL, *tracefp = NULL;
/* interpolation weights */
size_t num_interpolation_weights;
/* output usage */
if (argc <= 1)
usage();
/* initialize hts_engine API */
HTS_Engine_initialize(&engine);
/* get HTS voice file names */
num_voices = 0;
fn_voices = (char **) malloc(argc * sizeof(char *));
for (i = 0; i < argc; i++)
{
if (argv[i][0] == '-' && argv[i][1] == 'm')
fn_voices[num_voices++] = argv[++i];
if (argv[i][0] == '-' && argv[i][1] == 'h')
usage();
}
if (num_voices == 0)
{
fprintf(stderr, "Error: HTS voice must be specified.\n");
free(fn_voices);
return (-1);
}
/* load HTS voices */
if (HTS_Engine_load(&engine, fn_voices, num_voices) != TRUE)
{
fprintf(stderr, "Error: HTS voices cannot be loaded.\n");
free(fn_voices);
HTS_Engine_clear(&engine);
return (-1);
}
free(fn_voices);
/* get options */
while (--argc)
{
if (**++argv == '-')
{
switch (*(*argv + 1))
{
case 'v':
switch (*(*argv + 2))
{
case 'p':
HTS_Engine_set_phoneme_alignment_flag(&engine, TRUE);
break;
default:
fprintf(stderr, "Error: Invalid option '-v%c'.\n", *(*argv + 2));
HTS_Engine_clear(&engine);
return (-1);
}
break;
case 'o':
switch (*(*argv + 2))
{
case 'w':
wavfp = fopen(*++argv, "wb");
break;
case 'r':
rawfp = fopen(*++argv, "wb");
break;
case 'd':
durfp = fopen(*++argv, "wt");
break;
case 'm':
mgcfp = fopen(*++argv, "wb");
break;
case 'f':
case 'p':
lf0fp = fopen(*++argv, "wb");
break;
case 'l':
lpffp = fopen(*++argv, "wb");
break;
case 't':
tracefp = fopen(*++argv, "wt");
break;
default:
fprintf(stderr, "Error: Invalid option '-o%c'.\n", *(*argv + 2));
HTS_Engine_clear(&engine);
return (-1);
}
--argc;
break;
case 'h':
usage();
break;
case 'm':
argv++; /* HTS voices were already loaded */
--argc;
break;
case 's':
HTS_Engine_set_sampling_frequency(&engine, (size_t) atoi(*++argv));
--argc;
break;
case 'p':
HTS_Engine_set_fperiod(&engine, (size_t) atoi(*++argv));
--argc;
break;
case 'a':
HTS_Engine_set_alpha(&engine, atof(*++argv));
--argc;
break;
case 'b':
HTS_Engine_set_beta(&engine, atof(*++argv));
--argc;
break;
case 'r':
HTS_Engine_set_speed(&engine, atof(*++argv));
--argc;
break;
case 'f':
switch (*(*argv + 2))
{
case 'm':
HTS_Engine_add_half_tone(&engine, atof(*++argv));
break;
default:
fprintf(stderr, "Error: Invalid option '-f%c'.\n", *(*argv + 2));
HTS_Engine_clear(&engine);
return (-1);
}
--argc;
break;
case 'u':
HTS_Engine_set_msd_threshold(&engine, 1, atof(*++argv));
--argc;
break;
case 'i':
num_interpolation_weights = atoi(*++argv);
argc--;
if (num_interpolation_weights != num_voices)
{
HTS_Engine_clear(&engine);
return(-1);
}
for (i = 0; i < (int) num_interpolation_weights; i++)
{
f = atof(*++argv);
argc--;
HTS_Engine_set_duration_interpolation_weight(&engine, i, f);
HTS_Engine_set_parameter_interpolation_weight(&engine, i, 0, f);
HTS_Engine_set_parameter_interpolation_weight(&engine, i, 1, f);
HTS_Engine_set_gv_interpolation_weight(&engine, i, 0, f);
HTS_Engine_set_gv_interpolation_weight(&engine, i, 1, f);
}
break;
case 'j':
switch (*(*argv + 2))
{
case 'm':
HTS_Engine_set_gv_weight(&engine, 0, atof(*++argv));
break;
case 'f':
case 'p':
HTS_Engine_set_gv_weight(&engine, 1, atof(*++argv));
break;
default:
fprintf(stderr, "Error: Invalid option '-j%c'.\n", *(*argv + 2));
HTS_Engine_clear(&engine);
return(-1);
}
--argc;
break;
case 'g':
HTS_Engine_set_volume(&engine, atof(*++argv));
--argc;
break;
case 'z':
HTS_Engine_set_audio_buff_size(&engine, (size_t) atoi(*++argv));
--argc;
break;
default:
fprintf(stderr, "Error: Invalid option '-%c'.\n", *(*argv + 1));
HTS_Engine_clear(&engine);
return(-1);
}
}
else
{
labfn = *argv;
}
}
/* synthesize */
if (HTS_Engine_synthesize_from_fn(&engine, labfn) != TRUE)
{
fprintf(stderr, "Error: waveform cannot be synthesized.\n");
HTS_Engine_clear(&engine);
return(-1);
}
/* output */
if (tracefp != NULL)
HTS_Engine_save_information(&engine, tracefp);
if (durfp != NULL)
HTS_Engine_save_label(&engine, durfp);
if (rawfp)
HTS_Engine_save_generated_speech(&engine, rawfp);
if (wavfp)
HTS_Engine_save_riff(&engine, wavfp);
if (mgcfp)
HTS_Engine_save_generated_parameter(&engine, 0, mgcfp);
if (lf0fp)
HTS_Engine_save_generated_parameter(&engine, 1, lf0fp);
if (lpffp)
HTS_Engine_save_generated_parameter(&engine, 2, lpffp);
/* reset */
HTS_Engine_refresh(&engine);
/* free memory */
HTS_Engine_clear(&engine);
/* close files */
if (durfp != NULL)
fclose(durfp);
if (mgcfp != NULL)
fclose(mgcfp);
if (lf0fp != NULL)
fclose(lf0fp);
if (lpffp != NULL)
fclose(lpffp);
if (wavfp != NULL)
fclose(wavfp);
if (rawfp != NULL)
fclose(rawfp);
if (tracefp != NULL)
fclose(tracefp);
return 0;
}
int main(int argc, char **argv)
{
int i;
double f;
char *labfn = NULL;
size_t labfn_size=0;
char *guifn = NULL;
FILE *guifp = NULL;
FILE *rawfp = NULL;
/* number of speakers for interpolation */
int num_interp = 0;
double *rate_interp = NULL;
/* file names of models */
char **fn_ms_dur;
char **fn_ms_mgc;
char **fn_ms_lf0;
char **fn_ms_lpf;
/* number of each models for interpolation */
int num_ms_dur = 0, num_ms_mgc = 0, num_ms_lf0 = 0, num_ms_lpf = 0;
/* file names of trees */
char **fn_ts_dur;
char **fn_ts_mgc;
char **fn_ts_lf0;
char **fn_ts_lpf;
/* number of each trees for interpolation */
int num_ts_dur = 0, num_ts_mgc = 0, num_ts_lf0 = 0, num_ts_lpf = 0;
/* file names of windows */
char **fn_ws_mgc;
char **fn_ws_lf0;
char **fn_ws_lpf;
int num_ws_mgc = 0, num_ws_lf0 = 0, num_ws_lpf = 0;
/* file names of global variance */
char **fn_ms_gvm = NULL;
char **fn_ms_gvl = NULL;
char **fn_ms_gvf = NULL;
int num_ms_gvm = 0, num_ms_gvl = 0, num_ms_gvf = 0;
/* file names of global variance trees */
char **fn_ts_gvm = NULL;
char **fn_ts_gvl = NULL;
char **fn_ts_gvf = NULL;
int num_ts_gvm = 0, num_ts_gvl = 0, num_ts_gvf = 0;
/* file name of global variance switch */
char *fn_gv_switch = NULL;
/* global parameter */
int sampling_rate = 16000;
int fperiod = 80;
double alpha = 0.42;
int stage = 0; /* Gamma=-1/stage: if stage=0 then Gamma=0 */
double beta = 0.0;
int audio_buff_size = 1600;
double uv_threshold = 0.5;
double gv_weight_mgc = 1.0;
double gv_weight_lf0 = 1.0;
double gv_weight_lpf = 1.0;
double half_tone = 0.0;
HTS_Boolean phoneme_alignment = FALSE;
double speech_speed = 1.0;
HTS_Boolean use_log_gain = FALSE;
/* engine */
HTS_Engine engine;
/* parse command line */
if (argc == 1)
Usage();
/* delta window handler for mel-cepstrum */
fn_ws_mgc = (char **) calloc(argc, sizeof(char *));
/* delta window handler for log f0 */
fn_ws_lf0 = (char **) calloc(argc, sizeof(char *));
/* delta window handler for low-pass filter */
fn_ws_lpf = (char **) calloc(argc, sizeof(char *));
/* prepare for interpolation */
num_interp = GetNumInterp(argc, argv);
rate_interp = (double *) calloc(num_interp, sizeof(double));
for (i = 0; i < num_interp; i++)
rate_interp[i] = 1.0;
fn_ms_dur = (char **) calloc(num_interp, sizeof(char *));
fn_ms_mgc = (char **) calloc(num_interp, sizeof(char *));
fn_ms_lf0 = (char **) calloc(num_interp, sizeof(char *));
fn_ms_lpf = (char **) calloc(num_interp, sizeof(char *));
fn_ts_dur = (char **) calloc(num_interp, sizeof(char *));
fn_ts_mgc = (char **) calloc(num_interp, sizeof(char *));
fn_ts_lf0 = (char **) calloc(num_interp, sizeof(char *));
fn_ts_lpf = (char **) calloc(num_interp, sizeof(char *));
fn_ms_gvm = (char **) calloc(num_interp, sizeof(char *));
fn_ms_gvl = (char **) calloc(num_interp, sizeof(char *));
fn_ms_gvf = (char **) calloc(num_interp, sizeof(char *));
fn_ts_gvm = (char **) calloc(num_interp, sizeof(char *));
fn_ts_gvl = (char **) calloc(num_interp, sizeof(char *));
fn_ts_gvf = (char **) calloc(num_interp, sizeof(char *));
/* read command */
while (--argc) {
if (**++argv == '-') {
switch (*(*argv + 1)) {
case 'v':
switch (*(*argv + 2)) {
case 'p':
phoneme_alignment = TRUE;
break;
default:
Error(1, "hts_engine: Invalid option '-v%c'.\n", *(*argv + 2));
}
break;
case 't':
switch (*(*argv + 2)) {
case 'd':
fn_ts_dur[num_ts_dur++] = *++argv;
break;
case 'm':
fn_ts_mgc[num_ts_mgc++] = *++argv;
break;
case 'f':
case 'p':
fn_ts_lf0[num_ts_lf0++] = *++argv;
break;
case 'l':
fn_ts_lpf[num_ts_lpf++] = *++argv;
break;
default:
Error(1, "hts_engine: Invalid option '-t%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'm':
switch (*(*argv + 2)) {
case 'd':
fn_ms_dur[num_ms_dur++] = *++argv;
break;
case 'm':
fn_ms_mgc[num_ms_mgc++] = *++argv;
break;
case 'f':
case 'p':
fn_ms_lf0[num_ms_lf0++] = *++argv;
break;
case 'l':
fn_ms_lpf[num_ms_lpf++] = *++argv;
break;
default:
Error(1, "hts_engine: Invalid option '-m%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'd':
switch (*(*argv + 2)) {
case 'm':
fn_ws_mgc[num_ws_mgc++] = *++argv;
break;
case 'f':
case 'p':
fn_ws_lf0[num_ws_lf0++] = *++argv;
break;
case 'l':
fn_ws_lpf[num_ws_lpf++] = *++argv;
break;
default:
Error(1, "hts_engine: Invalid option '-d%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'o':
switch (*(*argv + 2)) {
case 'r':
rawfp = Getfp(*++argv, "ab");
break;
default:
Error(1, "festcat-hts_engine: Invalid option '-o%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'h':
Usage();
break;
case 's':
sampling_rate = atoi(*++argv);
--argc;
break;
case 'p':
fperiod = atoi(*++argv);
--argc;
break;
case 'a':
alpha = atof(*++argv);
--argc;
break;
case 'g':
stage = atoi(*++argv);
--argc;
break;
case 'l':
use_log_gain = TRUE;
break;
case 'b':
beta = atof(*++argv);
--argc;
break;
case 'r':
speech_speed = atof(*++argv);
--argc;
break;
case 'f':
switch (*(*argv + 2)) {
case 'm':
f = atof(*++argv);
if (f < -24.0)
f = -24.0;
if (f > 24.0)
f = 24.0;
half_tone = f;
break;
default:
Error(1, "hts_engine: Invalid option '-f%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'u':
uv_threshold = atof(*++argv);
--argc;
break;
case 'i':
++argv;
argc--;
for (i = 0; i < num_interp; i++) {
rate_interp[i] = atof(*++argv);
argc--;
}
break;
case 'e':
switch (*(*argv + 2)) {
case 'm':
fn_ts_gvm[num_ts_gvm++] = *++argv;
break;
case 'f':
case 'p':
fn_ts_gvl[num_ts_gvl++] = *++argv;
break;
case 'l':
fn_ts_gvf[num_ts_gvf++] = *++argv;
break;
default:
Error(1, "hts_engine: Invalid option '-e%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'c':
switch (*(*argv + 2)) {
case 'm':
fn_ms_gvm[num_ms_gvm++] = *++argv;
break;
case 'f':
case 'p':
fn_ms_gvl[num_ms_gvl++] = *++argv;
break;
case 'l':
fn_ms_gvf[num_ms_gvf++] = *++argv;
break;
default:
Error(1, "hts_engine: Invalid option '-c%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'j':
switch (*(*argv + 2)) {
case 'm':
gv_weight_mgc = atof(*++argv);
break;
case 'f':
case 'p':
gv_weight_lf0 = atof(*++argv);
break;
case 'l':
gv_weight_lpf = atof(*++argv);
break;
default:
Error(1, "hts_engine: Invalid option '-j%c'.\n", *(*argv + 2));
}
--argc;
break;
case 'k':
fn_gv_switch = *++argv;
--argc;
break;
case 'z':
audio_buff_size = atoi(*++argv);
--argc;
break;
default:
Error(1, "hts_engine: Invalid option '-%c'.\n", *(*argv + 1));
}
} else {
guifn = *argv;
}
}
/* number of models,trees check */
if (num_interp != num_ts_dur || num_interp != num_ts_mgc ||
num_interp != num_ts_lf0 || num_interp != num_ms_dur ||
num_interp != num_ms_mgc || num_interp != num_ms_lf0) {
Error(1, "hts_engine: specify %d models(trees) for each parameter.\n",
num_interp);
}
if (num_ms_lpf > 0 || num_ts_lpf > 0) {
if (num_interp != num_ms_lpf || num_interp != num_ts_lpf) {
Error(1, "hts_engine: specify %d models(trees) for each parameter.\n",
num_interp);
}
}
/* initialize (stream[0] = spectrum, stream[1] = lf0, stream[2] = low-pass filter) */
if (num_ms_lpf > 0 || num_ts_lpf > 0) {
HTS_Engine_initialize(&engine, 3);
} else {
HTS_Engine_initialize(&engine, 2);
}
/* load duration model */
HTS_Engine_load_duration_from_fn(&engine, fn_ms_dur, fn_ts_dur, num_interp);
/* load stream[0] (spectrum model) */
HTS_Engine_load_parameter_from_fn(&engine, fn_ms_mgc, fn_ts_mgc, fn_ws_mgc,
0, FALSE, num_ws_mgc, num_interp);
/* load stream[1] (lf0 model) */
HTS_Engine_load_parameter_from_fn(&engine, fn_ms_lf0, fn_ts_lf0, fn_ws_lf0,
1, TRUE, num_ws_lf0, num_interp);
/* load stream[2] (low-pass filter model) */
if (num_ms_lpf > 0 || num_ts_lpf > 0)
HTS_Engine_load_parameter_from_fn(&engine, fn_ms_lpf, fn_ts_lpf,
fn_ws_lpf, 2, FALSE, num_ws_lpf,
num_interp);
/* load gv[0] (GV for spectrum) */
if (num_interp == num_ms_gvm) {
if (num_ms_gvm == num_ts_gvm)
HTS_Engine_load_gv_from_fn(&engine, fn_ms_gvm, fn_ts_gvm, 0,
num_interp);
else
HTS_Engine_load_gv_from_fn(&engine, fn_ms_gvm, NULL, 0, num_interp);
}
/* load gv[1] (GV for lf0) */
if (num_interp == num_ms_gvl) {
if (num_ms_gvl == num_ts_gvl)
HTS_Engine_load_gv_from_fn(&engine, fn_ms_gvl, fn_ts_gvl, 1,
num_interp);
else
HTS_Engine_load_gv_from_fn(&engine, fn_ms_gvl, NULL, 1, num_interp);
}
/* load gv[2] (GV for low-pass filter) */
if (num_interp == num_ms_gvf && (num_ms_lpf > 0 || num_ts_lpf > 0)) {
if (num_ms_gvf == num_ts_gvf)
HTS_Engine_load_gv_from_fn(&engine, fn_ms_gvf, fn_ts_gvf, 0,
num_interp);
else
HTS_Engine_load_gv_from_fn(&engine, fn_ms_gvf, NULL, 2, num_interp);
}
/* load GV switch */
if (fn_gv_switch != NULL)
HTS_Engine_load_gv_switch_from_fn(&engine, fn_gv_switch);
/* set parameter */
HTS_Engine_set_sampling_rate(&engine, sampling_rate);
HTS_Engine_set_fperiod(&engine, fperiod);
HTS_Engine_set_alpha(&engine, alpha);
HTS_Engine_set_gamma(&engine, stage);
HTS_Engine_set_log_gain(&engine, use_log_gain);
HTS_Engine_set_beta(&engine, beta);
HTS_Engine_set_audio_buff_size(&engine, audio_buff_size);
HTS_Engine_set_msd_threshold(&engine, 1, uv_threshold); /* set voiced/unvoiced threshold for stream[1] */
HTS_Engine_set_gv_weight(&engine, 0, gv_weight_mgc);
HTS_Engine_set_gv_weight(&engine, 1, gv_weight_lf0);
if (num_ms_lpf > 0 || num_ts_lpf > 0)
HTS_Engine_set_gv_weight(&engine, 2, gv_weight_lpf);
for (i = 0; i < num_interp; i++) {
HTS_Engine_set_duration_interpolation_weight(&engine, i, rate_interp[i]);
HTS_Engine_set_parameter_interpolation_weight(&engine, 0, i,
rate_interp[i]);
HTS_Engine_set_parameter_interpolation_weight(&engine, 1, i,
rate_interp[i]);
if (num_ms_lpf > 0 || num_ts_lpf > 0)
HTS_Engine_set_parameter_interpolation_weight(&engine, 2, i,
rate_interp[i]);
}
if (num_interp == num_ms_gvm)
for (i = 0; i < num_interp; i++)
HTS_Engine_set_gv_interpolation_weight(&engine, 0, i, rate_interp[i]);
if (num_interp == num_ms_gvl)
for (i = 0; i < num_interp; i++)
HTS_Engine_set_gv_interpolation_weight(&engine, 1, i, rate_interp[i]);
if (num_interp == num_ms_gvf && (num_ms_lpf > 0 || num_ts_lpf > 0))
for (i = 0; i < num_interp; i++)
HTS_Engine_set_gv_interpolation_weight(&engine, 2, i, rate_interp[i]);
/* synthesis */
guifp=Getfp(guifn, "r");
while (getline(&labfn,&labfn_size,guifp) != -1)
{
chomp(labfn);
HTS_Engine_load_label_from_fn(&engine, labfn); /* load label file */
if (phoneme_alignment) /* modify label */
HTS_Label_set_frame_specified_flag(&engine.label, TRUE);
if (speech_speed != 1.0) /* modify label */
HTS_Label_set_speech_speed(&engine.label, speech_speed);
HTS_Engine_create_sstream(&engine); /* parse label and determine state duration */
if (half_tone != 0.0) { /* modify f0 */
for (i = 0; i < HTS_SStreamSet_get_total_state(&engine.sss); i++) {
f = HTS_SStreamSet_get_mean(&engine.sss, 1, i, 0);
f += half_tone * log(2.0) / 12;
if (f < log(10.0))
f = log(10.0);
HTS_SStreamSet_set_mean(&engine.sss, 1, i, 0, f);
}
}
HTS_Engine_create_pstream(&engine); /* generate speech parameter vector sequence */
HTS_Engine_create_gstream(&engine); /* synthesize speech */
/* output */
if (rawfp)
HTS_Engine_save_generated_speech(&engine, rawfp);
/* free */
HTS_Engine_refresh(&engine);
}
if (guifp != NULL)
fclose(guifp);
free(labfn);
/* free memory */
HTS_Engine_clear(&engine);
free(rate_interp);
free(fn_ws_mgc);
free(fn_ws_lf0);
free(fn_ws_lpf);
free(fn_ms_mgc);
free(fn_ms_lf0);
free(fn_ms_lpf);
free(fn_ms_dur);
free(fn_ts_mgc);
free(fn_ts_lf0);
free(fn_ts_lpf);
free(fn_ts_dur);
free(fn_ms_gvm);
free(fn_ms_gvl);
free(fn_ms_gvf);
free(fn_ts_gvm);
free(fn_ts_gvl);
free(fn_ts_gvf);
/* close files */
if (rawfp != NULL)
fclose(rawfp);
return 0;
}
