/* HTS_Engine_save_generated_speech: output generated speech */
void HTS_Engine_save_generated_speech(HTS_Engine * engine, FILE * fp)
{
int i;
short temp;
HTS_GStreamSet *gss = &engine->gss;
for (i = 0; i < HTS_GStreamSet_get_total_nsample(gss); i++) {
temp = HTS_GStreamSet_get_speech(gss, i);
fwrite(&temp, sizeof(short), 1, fp);
}
}
/* HTS_Engine_save_riff: save RIFF format file */
void HTS_Engine_save_riff(HTS_Engine * engine, FILE * fp)
{
size_t i;
double x;
short temp;
HTS_GStreamSet *gss = &engine->gss;
char data_01_04[] = { 'R', 'I', 'F', 'F' };
int data_05_08 = HTS_GStreamSet_get_total_nsamples(gss) * sizeof(short) + 36;
char data_09_12[] = { 'W', 'A', 'V', 'E' };
char data_13_16[] = { 'f', 'm', 't', ' ' };
int data_17_20 = 16;
short data_21_22 = 1; /* PCM */
short data_23_24 = 1; /* monoral */
int data_25_28 = engine->condition.sampling_frequency;
int data_29_32 = engine->condition.sampling_frequency * sizeof(short);
short data_33_34 = sizeof(short);
short data_35_36 = (short) (sizeof(short) * 8);
char data_37_40[] = { 'd', 'a', 't', 'a' };
int data_41_44 = HTS_GStreamSet_get_total_nsamples(gss) * sizeof(short);
/* write header */
HTS_fwrite_little_endian(data_01_04, sizeof(char), 4, fp);
HTS_fwrite_little_endian(&data_05_08, sizeof(int), 1, fp);
HTS_fwrite_little_endian(data_09_12, sizeof(char), 4, fp);
HTS_fwrite_little_endian(data_13_16, sizeof(char), 4, fp);
HTS_fwrite_little_endian(&data_17_20, sizeof(int), 1, fp);
HTS_fwrite_little_endian(&data_21_22, sizeof(short), 1, fp);
HTS_fwrite_little_endian(&data_23_24, sizeof(short), 1, fp);
HTS_fwrite_little_endian(&data_25_28, sizeof(int), 1, fp);
HTS_fwrite_little_endian(&data_29_32, sizeof(int), 1, fp);
HTS_fwrite_little_endian(&data_33_34, sizeof(short), 1, fp);
HTS_fwrite_little_endian(&data_35_36, sizeof(short), 1, fp);
HTS_fwrite_little_endian(data_37_40, sizeof(char), 4, fp);
HTS_fwrite_little_endian(&data_41_44, sizeof(int), 1, fp);
/* write data */
for (i = 0; i < HTS_GStreamSet_get_total_nsamples(gss); i++) {
x = HTS_GStreamSet_get_speech(gss, i);
if (x > 32767.0)
temp = 32767;
else if (x < -32768.0)
temp = -32768;
else
temp = (short) x;
HTS_fwrite_little_endian(&temp, sizeof(short), 1, fp);
}
}
/* HTS_Engine_save_generated_speech: save generated speech */
void HTS_Engine_save_generated_speech(HTS_Engine * engine, FILE * fp)
{
size_t i;
double x;
short temp;
HTS_GStreamSet *gss = &engine->gss;
for (i = 0; i < HTS_GStreamSet_get_total_nsamples(gss); i++) {
x = HTS_GStreamSet_get_speech(gss, i);
if (x > 32767.0)
temp = 32767;
else if (x < -32768.0)
temp = -32768;
else
temp = (short) x;
fwrite(&temp, sizeof(short), 1, fp);
}
}
/* HTS_Engine_save_riff: output RIFF format file */
void HTS_Engine_save_riff(HTS_Engine * engine, FILE * fp)
{
int i;
short temp;
HTS_GStreamSet *gss = &engine->gss;
char data_01_04[] = { 'R', 'I', 'F', 'F' };
int data_05_08 = HTS_GStreamSet_get_total_nsample(gss) * sizeof(short) + 36;
char data_09_12[] = { 'W', 'A', 'V', 'E' };
char data_13_16[] = { 'f', 'm', 't', ' ' };
int data_17_20 = 16;
short data_21_22 = 1; /* PCM */
short data_23_24 = 1; /* monoral */
int data_25_28 = engine->global.sampling_rate;
int data_29_32 = engine->global.sampling_rate * sizeof(short);
short data_33_34 = sizeof(short);
short data_35_36 = (short) (sizeof(short) * 8);
char data_37_40[] = { 'd', 'a', 't', 'a' };
int data_41_44 = HTS_GStreamSet_get_total_nsample(gss) * sizeof(short);
/* write header */
HTS_fwrite_little_endian(data_01_04, sizeof(char), 4, fp);
HTS_fwrite_little_endian(&data_05_08, sizeof(int), 1, fp);
HTS_fwrite_little_endian(data_09_12, sizeof(char), 4, fp);
HTS_fwrite_little_endian(data_13_16, sizeof(char), 4, fp);
HTS_fwrite_little_endian(&data_17_20, sizeof(int), 1, fp);
HTS_fwrite_little_endian(&data_21_22, sizeof(short), 1, fp);
HTS_fwrite_little_endian(&data_23_24, sizeof(short), 1, fp);
HTS_fwrite_little_endian(&data_25_28, sizeof(int), 1, fp);
HTS_fwrite_little_endian(&data_29_32, sizeof(int), 1, fp);
HTS_fwrite_little_endian(&data_33_34, sizeof(short), 1, fp);
HTS_fwrite_little_endian(&data_35_36, sizeof(short), 1, fp);
HTS_fwrite_little_endian(data_37_40, sizeof(char), 4, fp);
HTS_fwrite_little_endian(&data_41_44, sizeof(int), 1, fp);
/* write data */
for (i = 0; i < HTS_GStreamSet_get_total_nsample(gss); i++) {
temp = HTS_GStreamSet_get_speech(gss, i);
HTS_fwrite_little_endian(&temp, sizeof(short), 1, fp);
}
}
/* HTS_Engine_get_generated_speech: output generated speech */
double HTS_Engine_get_generated_speech(HTS_Engine * engine, size_t index)
{
return HTS_GStreamSet_get_speech(&engine->gss, index);
}
static void Run(const SUttProcessor *self, SUtterance *utt,
s_erc *error)
{
SHTSEngineSynthUttProc104 *HTSsynth = (SHTSEngineSynthUttProc104*)self;
SPlugin *audioPlugin;
const SRelation *segmentRel;
SAudio *audio = NULL;
s_bool is_present;
char **label_data = NULL;
int label_size;
const SItem *item;
const SItem *itemItr;
int counter;
uint i;
int frame;
int state;
S_CLR_ERR(error);
/* we require the segment relation */
is_present = SUtteranceRelationIsPresent(utt, "Segment", error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SUtteranceRelationIsPresent\" failed"))
goto quit_error;
if (!is_present)
{
S_CTX_ERR(error, S_FAILURE,
"Run",
"Failed to find 'Segment' relation in utterance");
goto quit_error;
}
segmentRel = SUtteranceGetRelation(utt, "Segment", error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SUtteranceGetRelation\" failed"))
goto quit_error;
item = SRelationHead(segmentRel, error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SRelationHead\" failed"))
goto quit_error;
itemItr = item;
label_size = 0;
while (itemItr != NULL)
{
label_size++;
itemItr = SItemNext(itemItr, error);
}
label_data = S_CALLOC(char*, label_size);
itemItr = item;
counter = 0;
while (itemItr != NULL)
{
SObject *dFeat;
const char *tmp;
dFeat = SItemPathToFeatProc(itemItr, "hts_labels", error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SItemPathToFeatProc\" failed"))
goto quit_error;
if (dFeat == NULL)
{
S_CTX_ERR(error, S_FAILURE,
"Run",
"Failed to generate hts labels for segment item");
goto quit_error;
}
tmp = SObjectGetString(dFeat, error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SObjectGetString\" failed"))
goto quit_error;
label_data[counter++] = s_strdup(tmp, error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"s_strdup\" failed"))
goto quit_error;
SItemSetObject((SItem*)itemItr, "hts_label", dFeat, error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SItemSetObject\" failed"))
goto quit_error;
itemItr = SItemNext(itemItr, error);
}
/* speech synthesis part */
HTS_Engine_load_label_from_string_list(&(HTSsynth->engine), label_data, label_size);
HTS_Engine_create_sstream(&(HTSsynth->engine));
HTS_Engine_create_pstream(&(HTSsynth->engine));
HTS_Engine_create_gstream(&(HTSsynth->engine));
itemItr = item;
counter = 0;
frame = 0;
state = 0;
while (itemItr != NULL)
{
int j;
int duration;
HTS_SStreamSet *sss = &(HTSsynth->engine.sss);
const int nstate = HTS_ModelSet_get_nstate(&(HTSsynth->engine.ms));
const double rate = HTSsynth->engine.global.fperiod * 1e+7 / HTSsynth->engine.global.sampling_rate;
float tmp;
for (j = 0, duration = 0; j < nstate; j++)
duration += HTS_SStreamSet_get_duration(sss, state++);
tmp = frame * rate;
SItemSetFloat((SItem*)itemItr, "start", tmp/1e+7, error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SItemSetFloat\" failed"))
goto quit_error;
tmp = (frame + duration) * rate;
SItemSetFloat((SItem*)itemItr, "end", tmp/1e+7, error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SItemSetFloat\" failed"))
goto quit_error;
frame += duration;
itemItr = SItemNext(itemItr, error);
counter++;
}
/* create an audio object */
audio = S_NEW(SAudio, error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Failed to create new 'SAudio' object"))
goto quit_error;
/* set audio feature in utterance */
SUtteranceSetFeature(utt, "audio", S_OBJECT(audio), error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SUtteranceSetFeature\" failed"))
goto quit_error;
/* We need to give the utterance the audio plug-in. If we don't do
* this and the voice is deleted before the utterance, then the
* utterance can't do *anything* with the audio. Not even delete
* it (segfault). This should be fast because it is already
* loaded.
* Note that this happens after the audio is set. This is because
* utt features are a list implementation.
*/
audioPlugin = s_pm_load_plugin("audio.spi", error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SUtteranceSetFeature\" failed"))
goto quit_error;
SUtteranceSetFeature(utt, "audio_plugin", S_OBJECT(audioPlugin), error);
if (S_CHK_ERR(error, S_CONTERR,
"Run",
"Call to \"SUtteranceSetFeature\" failed"))
{
S_DELETE(audioPlugin, "Run", error);
goto quit_error;
}
audio->sample_rate = HTSsynth->engine.global.sampling_rate;
audio->num_samples = (uint32)HTS_GStreamSet_get_total_nsample(&(HTSsynth->engine).gss);
audio->samples = S_MALLOC(float, audio->num_samples);
if (audio->samples == NULL)
{
S_FTL_ERR(error, S_MEMERROR,
"Run",
"Failed to allocate memory for 'float' object");
goto quit_error;
}
/* write data */
for (i = 0; i < audio->num_samples; i++)
audio->samples[i] = (float)(HTS_GStreamSet_get_speech(&(HTSsynth->engine).gss, i) * 1.0);
for (counter = 0; counter < label_size; counter++)
S_FREE(label_data[counter]);
S_FREE(label_data);
HTS_Engine_refresh(&(HTSsynth->engine));
/* all OK here */
return;
/* error clean-up code */
quit_error:
if (label_data != NULL)
{
for (counter = 0; counter < label_size; counter++)
{
if (label_data[counter] != NULL)
S_FREE(label_data[counter]);
}
S_FREE(label_data);
}
return;
}
