/* HTS_Engine_clear: free engine */
void HTS_Engine_clear(HTS_Engine * engine)
{
size_t i;
if (engine->condition.msd_threshold != NULL)
HTS_free(engine->condition.msd_threshold);
if (engine->condition.duration_iw != NULL)
HTS_free(engine->condition.duration_iw);
if (engine->condition.gv_weight != NULL)
HTS_free(engine->condition.gv_weight);
if (engine->condition.parameter_iw != NULL) {
for (i = 0; i < HTS_ModelSet_get_nvoices(&engine->ms); i++)
HTS_free(engine->condition.parameter_iw[i]);
HTS_free(engine->condition.parameter_iw);
}
if (engine->condition.gv_iw != NULL) {
for (i = 0; i < HTS_ModelSet_get_nvoices(&engine->ms); i++)
HTS_free(engine->condition.gv_iw[i]);
HTS_free(engine->condition.gv_iw);
}
HTS_ModelSet_clear(&engine->ms);
HTS_Audio_clear(&engine->audio);
HTS_Engine_initialize(engine);
}
/* HTS_Engine_get_nvoices: get number of voices */
size_t HTS_Engine_get_nvoices(HTS_Engine * engine)
{
return HTS_ModelSet_get_nvoices(&engine->ms);
}
/* HTS_Engine_save_information: save trace information */
void HTS_Engine_save_information(HTS_Engine * engine, FILE * fp)
{
size_t i, j, k, l, m, n;
double temp;
HTS_Condition *condition = &engine->condition;
HTS_ModelSet *ms = &engine->ms;
HTS_Label *label = &engine->label;
HTS_SStreamSet *sss = &engine->sss;
HTS_PStreamSet *pss = &engine->pss;
/* global parameter */
fprintf(fp, "[Global parameter]\n");
fprintf(fp, "Sampring frequency -> %8lu(Hz)\n", (unsigned long) condition->sampling_frequency);
fprintf(fp, "Frame period -> %8lu(point)\n", (unsigned long) condition->fperiod);
fprintf(fp, " %8.5f(msec)\n", 1e+3 * condition->fperiod / condition->sampling_frequency);
fprintf(fp, "All-pass constant -> %8.5f\n", (float) condition->alpha);
fprintf(fp, "Gamma -> %8.5f\n", (float) (condition->stage == 0 ? 0.0 : -1.0 / condition->stage));
if (condition->stage != 0) {
if (condition->use_log_gain == TRUE)
fprintf(fp, "Log gain flag -> TRUE\n");
else
fprintf(fp, "Log gain flag -> FALSE\n");
}
fprintf(fp, "Postfiltering coefficient -> %8.5f\n", (float) condition->beta);
fprintf(fp, "Audio buffer size -> %8lu(sample)\n", (unsigned long) condition->audio_buff_size);
fprintf(fp, "\n");
/* duration parameter */
fprintf(fp, "[Duration parameter]\n");
fprintf(fp, "Number of states -> %8lu\n", (unsigned long) HTS_ModelSet_get_nstate(ms));
fprintf(fp, " Interpolation size -> %8lu\n", (unsigned long) HTS_ModelSet_get_nvoices(ms));
/* check interpolation */
for (i = 0, temp = 0.0; i < HTS_ModelSet_get_nvoices(ms); i++)
temp += condition->duration_iw[i];
for (i = 0; i < HTS_ModelSet_get_nvoices(ms); i++)
if (condition->duration_iw[i] != 0.0)
condition->duration_iw[i] /= temp;
for (i = 0; i < HTS_ModelSet_get_nvoices(ms); i++)
fprintf(fp, " Interpolation weight[%2lu] -> %8.0f(%%)\n", (unsigned long) i, (float) (100 * condition->duration_iw[i]));
fprintf(fp, "\n");
fprintf(fp, "[Stream parameter]\n");
for (i = 0; i < HTS_ModelSet_get_nstream(ms); i++) {
/* stream parameter */
fprintf(fp, "Stream[%2lu] vector length -> %8lu\n", (unsigned long) i, (unsigned long) HTS_ModelSet_get_vector_length(ms, i));
fprintf(fp, " Dynamic window size -> %8lu\n", (unsigned long) HTS_ModelSet_get_window_size(ms, i));
/* interpolation */
fprintf(fp, " Interpolation size -> %8lu\n", (unsigned long) HTS_ModelSet_get_nvoices(ms));
for (j = 0, temp = 0.0; j < HTS_ModelSet_get_nvoices(ms); j++)
temp += condition->parameter_iw[j][i];
for (j = 0; j < HTS_ModelSet_get_nvoices(ms); j++)
if (condition->parameter_iw[j][i] != 0.0)
condition->parameter_iw[j][i] /= temp;
for (j = 0; j < HTS_ModelSet_get_nvoices(ms); j++)
fprintf(fp, " Interpolation weight[%2lu] -> %8.0f(%%)\n", (unsigned long) j, (float) (100 * condition->parameter_iw[j][i]));
/* MSD */
if (HTS_ModelSet_is_msd(ms, i)) { /* for MSD */
fprintf(fp, " MSD flag -> TRUE\n");
fprintf(fp, " MSD threshold -> %8.5f\n", condition->msd_threshold[i]);
} else { /* for non MSD */
fprintf(fp, " MSD flag -> FALSE\n");
}
/* GV */
if (HTS_ModelSet_use_gv(ms, i)) {
fprintf(fp, " GV flag -> TRUE\n");
fprintf(fp, " GV weight -> %8.0f(%%)\n", (float) (100 * condition->gv_weight[i]));
fprintf(fp, " GV interpolation size -> %8lu\n", (unsigned long) HTS_ModelSet_get_nvoices(ms));
/* interpolation */
for (j = 0, temp = 0.0; j < HTS_ModelSet_get_nvoices(ms); j++)
temp += condition->gv_iw[j][i];
for (j = 0; j < HTS_ModelSet_get_nvoices(ms); j++)
if (condition->gv_iw[j][i] != 0.0)
condition->gv_iw[j][i] /= temp;
for (j = 0; j < HTS_ModelSet_get_nvoices(ms); j++)
fprintf(fp, " GV interpolation weight[%2lu] -> %8.0f(%%)\n", (unsigned long) j, (float) (100 * condition->gv_iw[j][i]));
} else {
fprintf(fp, " GV flag -> FALSE\n");
}
}
fprintf(fp, "\n");
/* generated sequence */
fprintf(fp, "[Generated sequence]\n");
fprintf(fp, "Number of HMMs -> %8lu\n", (unsigned long) HTS_Label_get_size(label));
fprintf(fp, "Number of stats -> %8lu\n", (unsigned long) HTS_Label_get_size(label) * HTS_ModelSet_get_nstate(ms));
fprintf(fp, "Length of this speech -> %8.3f(sec)\n", (float) ((double) HTS_PStreamSet_get_total_frame(pss) * condition->fperiod / condition->sampling_frequency));
fprintf(fp, " -> %8lu(frames)\n", (unsigned long) HTS_PStreamSet_get_total_frame(pss) * condition->fperiod);
for (i = 0; i < HTS_Label_get_size(label); i++) {
fprintf(fp, "HMM[%2lu]\n", (unsigned long) i);
fprintf(fp, " Name -> %s\n", HTS_Label_get_string(label, i));
fprintf(fp, " Duration\n");
for (j = 0; j < HTS_ModelSet_get_nvoices(ms); j++) {
fprintf(fp, " Interpolation[%2lu]\n", (unsigned long) j);
HTS_ModelSet_get_duration_index(ms, j, HTS_Label_get_string(label, i), &k, &l);
fprintf(fp, " Tree index -> %8lu\n", (unsigned long) k);
fprintf(fp, " PDF index -> %8lu\n", (unsigned long) l);
}
for (j = 0; j < HTS_ModelSet_get_nstate(ms); j++) {
fprintf(fp, " State[%2lu]\n", (unsigned long) j + 2);
fprintf(fp, " Length -> %8lu(frames)\n", (unsigned long) HTS_SStreamSet_get_duration(sss, i * HTS_ModelSet_get_nstate(ms) + j));
for (k = 0; k < HTS_ModelSet_get_nstream(ms); k++) {
fprintf(fp, " Stream[%2lu]\n", (unsigned long) k);
if (HTS_ModelSet_is_msd(ms, k)) {
if (HTS_SStreamSet_get_msd(sss, k, i * HTS_ModelSet_get_nstate(ms) + j) > condition->msd_threshold[k])
fprintf(fp, " MSD flag -> TRUE\n");
else
fprintf(fp, " MSD flag -> FALSE\n");
}
for (l = 0; l < HTS_ModelSet_get_nvoices(ms); l++) {
fprintf(fp, " Interpolation[%2lu]\n", (unsigned long) l);
HTS_ModelSet_get_parameter_index(ms, l, k, j + 2, HTS_Label_get_string(label, i), &m, &n);
fprintf(fp, " Tree index -> %8lu\n", (unsigned long) m);
fprintf(fp, " PDF index -> %8lu\n", (unsigned long) n);
}
}
}
}
}
/* HTS_SStreamSet_create: parse label and determine state duration */
HTS_Boolean HTS_SStreamSet_create(HTS_SStreamSet * sss, HTS_ModelSet * ms, HTS_Label * label, HTS_Boolean phoneme_alignment_flag, double speed, double *duration_iw, double **parameter_iw, double **gv_iw)
{
size_t i, j, k;
double temp;
int shift;
size_t state;
HTS_SStream *sst;
double *duration_mean, *duration_vari;
double frame_length;
size_t next_time;
size_t next_state;
/* check interpolation weights */
for (i = 0, temp = 0.0; i < HTS_ModelSet_get_nvoices(ms); i++)
temp += duration_iw[i];
if (temp == 0.0) {
return FALSE;
} else if (temp != 1.0) {
for (i = 0; i < HTS_ModelSet_get_nvoices(ms); i++)
if (duration_iw[i] != 0.0)
duration_iw[i] /= temp;
}
for (i = 0; i < HTS_ModelSet_get_nstream(ms); i++) {
for (j = 0, temp = 0.0; j < HTS_ModelSet_get_nvoices(ms); j++)
temp += parameter_iw[i][j];
if (temp == 0.0) {
return FALSE;
} else if (temp != 1.0) {
for (j = 0; j < HTS_ModelSet_get_nvoices(ms); j++)
if (parameter_iw[i][j] != 0.0)
parameter_iw[i][j] /= temp;
}
if (HTS_ModelSet_use_gv(ms, i)) {
for (j = 0, temp = 0.0; j < HTS_ModelSet_get_nvoices(ms); j++)
temp += gv_iw[i][j];
if (temp == 0.0)
return FALSE;
else if (temp != 1.0)
for (j = 0; j < HTS_ModelSet_get_nvoices(ms); j++)
if (gv_iw[i][j] != 0.0)
gv_iw[i][j] /= temp;
}
}
/* initialize state sequence */
sss->nstate = HTS_ModelSet_get_nstate(ms);
sss->nstream = HTS_ModelSet_get_nstream(ms);
sss->total_frame = 0;
sss->total_state = HTS_Label_get_size(label) * sss->nstate;
sss->duration = (size_t *) HTS_calloc(sss->total_state, sizeof(size_t));
sss->sstream = (HTS_SStream *) HTS_calloc(sss->nstream, sizeof(HTS_SStream));
for (i = 0; i < sss->nstream; i++) {
sst = &sss->sstream[i];
sst->vector_length = HTS_ModelSet_get_vector_length(ms, i);
sst->mean = (double **) HTS_calloc(sss->total_state, sizeof(double *));
sst->vari = (double **) HTS_calloc(sss->total_state, sizeof(double *));
if (HTS_ModelSet_is_msd(ms, i))
sst->msd = (double *) HTS_calloc(sss->total_state, sizeof(double));
else
sst->msd = NULL;
for (j = 0; j < sss->total_state; j++) {
sst->mean[j] = (double *) HTS_calloc(sst->vector_length * HTS_ModelSet_get_window_size(ms, i), sizeof(double));
sst->vari[j] = (double *) HTS_calloc(sst->vector_length * HTS_ModelSet_get_window_size(ms, i), sizeof(double));
}
if (HTS_ModelSet_use_gv(ms, i)) {
sst->gv_switch = (HTS_Boolean *) HTS_calloc(sss->total_state, sizeof(HTS_Boolean));
for (j = 0; j < sss->total_state; j++)
sst->gv_switch[j] = TRUE;
} else {
sst->gv_switch = NULL;
}
}
/* determine state duration */
duration_mean = (double *) HTS_calloc(sss->total_state, sizeof(double));
duration_vari = (double *) HTS_calloc(sss->total_state, sizeof(double));
for (i = 0; i < HTS_Label_get_size(label); i++)
HTS_ModelSet_get_duration(ms, HTS_Label_get_string(label, i), duration_iw, &duration_mean[i * sss->nstate], &duration_vari[i * sss->nstate]);
if (phoneme_alignment_flag == TRUE) {
/* use duration set by user */
next_time = 0;
next_state = 0;
state = 0;
for (i = 0; i < HTS_Label_get_size(label); i++) {
temp = HTS_Label_get_end_frame(label, i);
if (temp >= 0) {
next_time += (size_t) HTS_set_specified_duration(&sss->duration[next_state], &duration_mean[next_state], &duration_vari[next_state], state + sss->nstate - next_state, temp - next_time);
next_state = state + sss->nstate;
} else if (i + 1 == HTS_Label_get_size(label)) {
HTS_error(-1, "HTS_SStreamSet_create: The time of final label is not specified.\n");
HTS_set_default_duration(&sss->duration[next_state], &duration_mean[next_state], &duration_vari[next_state], state + sss->nstate - next_state);
}
state += sss->nstate;
}
} else {
/* determine frame length */
if (speed != 1.0) {
temp = 0.0;
for (i = 0; i < sss->total_state; i++) {
temp += duration_mean[i];
}
frame_length = temp / speed;
HTS_set_specified_duration(sss->duration, duration_mean, duration_vari, sss->total_state, frame_length);
} else {
HTS_set_default_duration(sss->duration, duration_mean, duration_vari, sss->total_state);
}
}
HTS_free(duration_mean);
HTS_free(duration_vari);
/* get parameter */
for (i = 0, state = 0; i < HTS_Label_get_size(label); i++) {
for (j = 2; j <= sss->nstate + 1; j++) {
sss->total_frame += sss->duration[state];
for (k = 0; k < sss->nstream; k++) {
sst = &sss->sstream[k];
if (sst->msd)
HTS_ModelSet_get_parameter(ms, k, j, HTS_Label_get_string(label, i), parameter_iw[k], sst->mean[state], sst->vari[state], &sst->msd[state]);
else
HTS_ModelSet_get_parameter(ms, k, j, HTS_Label_get_string(label, i), parameter_iw[k], sst->mean[state], sst->vari[state], NULL);
}
state++;
}
}
/* copy dynamic window */
for (i = 0; i < sss->nstream; i++) {
sst = &sss->sstream[i];
sst->win_size = HTS_ModelSet_get_window_size(ms, i);
sst->win_max_width = HTS_ModelSet_get_window_max_width(ms, i);
sst->win_l_width = (int *) HTS_calloc(sst->win_size, sizeof(int));
sst->win_r_width = (int *) HTS_calloc(sst->win_size, sizeof(int));
sst->win_coefficient = (double **) HTS_calloc(sst->win_size, sizeof(double));
for (j = 0; j < sst->win_size; j++) {
sst->win_l_width[j] = HTS_ModelSet_get_window_left_width(ms, i, j);
sst->win_r_width[j] = HTS_ModelSet_get_window_right_width(ms, i, j);
if (sst->win_l_width[j] + sst->win_r_width[j] == 0)
sst->win_coefficient[j] = (double *) HTS_calloc(-2 * sst->win_l_width[j] + 1, sizeof(double));
else
sst->win_coefficient[j] = (double *) HTS_calloc(-2 * sst->win_l_width[j], sizeof(double));
sst->win_coefficient[j] -= sst->win_l_width[j];
for (shift = sst->win_l_width[j]; shift <= sst->win_r_width[j]; shift++)
sst->win_coefficient[j][shift] = HTS_ModelSet_get_window_coefficient(ms, i, j, shift);
}
}
/* determine GV */
for (i = 0; i < sss->nstream; i++) {
sst = &sss->sstream[i];
if (HTS_ModelSet_use_gv(ms, i)) {
sst->gv_mean = (double *) HTS_calloc(sst->vector_length, sizeof(double));
sst->gv_vari = (double *) HTS_calloc(sst->vector_length, sizeof(double));
HTS_ModelSet_get_gv(ms, i, HTS_Label_get_string(label, 0), gv_iw[i], sst->gv_mean, sst->gv_vari);
} else {
sst->gv_mean = NULL;
sst->gv_vari = NULL;
}
}
for (i = 0; i < HTS_Label_get_size(label); i++)
if (HTS_ModelSet_get_gv_flag(ms, HTS_Label_get_string(label, i)) == FALSE)
for (j = 0; j < sss->nstream; j++)
if (HTS_ModelSet_use_gv(ms, j) == TRUE)
for (k = 0; k < sss->nstate; k++)
sss->sstream[j].gv_switch[i * sss->nstate + k] = FALSE;
return TRUE;
}
