[hts-users:01455] gstream redundancy

["Nicholas Volk" <nvolk@xxxxxxxxxx>]

Hi,

Is there some reason for duplicating pstream data to gstream data?
Meaning that we could reduce memory consumption and time spend of making
duplicates by removing gstreams and writing HTS_GStreamSet_create() simply
as shown below. (I use pss->total_frame instead of the API function etc,
but that is trivial.)

br,
  Nicholas



/* HTS_GStreamSet_create: generate speech */
/* (stream[0] == spectrum && stream[1] == lf0) */
void HTS_GStreamSet_create(HTS_GStreamSet * gss, HTS_PStreamSet * pss,
                           int stage, int sampling_rate, int fperiod,
                           double alpha, double beta, int audio_buff_size) {
   int i, j, k;

   int msd_frame;
   HTS_Vocoder v;

   /* check */
   if ( gss->gspeech )
     HTS_error(1, "HTS_GStreamSet_create: HTS_GStreamSet has to
initialize.");

   /* initialize */
   gss->total_nsample = fperiod * pss->total_frame;

   /* check */
   if (pss->nstream != 2)
      HTS_error(1, "HTS_GStreamSet_create: nstream != 2 is not support.");
   if (pss->pstream[1].static_length != 1)
      HTS_error(1,
                "HTS_GStreamSet_create: lf0 static vector size != 1 is not
support.");

   gss->gspeech = (short *) HTS_calloc(gss->total_nsample, sizeof(short));

   /* synthesize speech waveform */
   HTS_Vocoder_initialize(&v, pss->pstream[0].static_length - 1, stage,
                          sampling_rate, fperiod, audio_buff_size);
   for (i = 0, msd_frame=0; i < pss->total_frame; i++) {
     double lf0 = LZERO;
     if ( pss->pstream[1].msd_flag[i] ) {
       lf0 = pss->pstream[1].par[msd_frame++][0];
     }
     HTS_Vocoder_synthesize(&v, pss->pstream[0].static_length - 1,
			    lf0,
			    &pss->pstream[0].par[i][0], alpha, beta,
			    &gss->gspeech[i * fperiod]);
   }
   HTS_Vocoder_clear(&v);
}

Ok, I use /* HTS_GStreamSet_create: generate speech */
/* (stream[0] == spectrum && stream[1] == lf0) */
void HTS_GStreamSet_create(HTS_GStreamSet * gss, HTS_PStreamSet * pss,
                           int stage, int sampling_rate, int fperiod,
                           double alpha, double beta, int audio_buff_size) {
   int i, j, k;

   int msd_frame;
   HTS_Vocoder v;

   /* check */
   if ( gss->gspeech )
     HTS_error(1, "HTS_GStreamSet_create: HTS_GStreamSet has to
initialize.");

   /* initialize */
   gss->total_nsample = fperiod * pss->total_frame;

   /* check */
   if (pss->nstream != 2)
      HTS_error(1, "HTS_GStreamSet_create: nstream != 2 is not support.");
   if (pss->pstream[1].static_length != 1)
      HTS_error(1,
                "HTS_GStreamSet_create: lf0 static vector size != 1 is not
support.");

   gss->gspeech = (short *) HTS_calloc(gss->total_nsample, sizeof(short));

   /* synthesize speech waveform */
   HTS_Vocoder_initialize(&v, pss->pstream[0].static_length - 1, stage,
                          sampling_rate, fperiod, audio_buff_size);
   for (i = 0, msd_frame=0; i < pss->total_frame; i++) {
     double lf0 = LZERO;
     if ( pss->pstream[1].msd_flag[i] ) {
       lf0 = pss->pstream[1].par[msd_frame++][0];
     }
     HTS_Vocoder_synthesize(&v, pss->pstream[0].static_length - 1,
			    lf0,
			    &pss->pstream[0].par[i][0], alpha, beta,
			    &gss->gspeech[i * fperiod]);
   }
   HTS_Vocoder_clear(&v);
}