wav2mfcc.c

Go to the documentation of this file.

/*
 * Copyright (c) 1991-2006 Kawahara Lab., Kyoto University
 * Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and Technology
 * Copyright (c) 2005-2006 Julius project team, Nagoya Institute of Technology, Nagoya Institute of Technology
 * All rights reserved
 */

#include <julius.h>

#include <sys/stat.h>

/* make MFCC_E_D_N_Z from speech[0..speechlen-1] and returns
   newly malloced param */

void
init_para(Value *para)
{
  para->smp_period = smpPeriod;
  para->framesize  = fsize;
  para->frameshift = fshift;
  para->preEmph    = preemph;
  para->mfcc_dim   = model_mfcc_dim;
  para->lifter     = ceplifter;
  para->delWin     = delwin;
  para->accWin     = accwin;
  para->hipass     = hipass;
  para->lopass     = lopass;
  para->c0         = c0_required;
  para->fbank_num  = fbank_num;
  para->ss_alpha   = ssalpha;
  para->ss_floor   = ssfloor;
  para->cmn        = cmn_required;
  para->raw_e      = rawe_required;
  para->zmeanframe = zmean_frame;
  para->enormal    = enormal_required;
  /* para->escale     = DEF_ESCALE; */
  para->escale     = enormal_escale;
  para->silFloor   = enormal_silfloor;

  para->delta = delta_required;
  para->acc   = acc_required;
  para->energy = energy_required;
  para->absesup = abs_energy_suppress;

  /* determine base size */
  para->baselen = para->mfcc_dim + (para->c0 ? 1 : 0) + (para->energy ? 1 : 0);
  /* determine maximim size of parameter vector for temporal buffer */
  para->vecbuflen = para->baselen * (1 + (para->delta ? 1 : 0) + (para->acc ? 1 : 0));
  /* determine final size of parameter vector */
  para->veclen = para->vecbuflen - (para->absesup ? 1 : 0);
}

HTK_Param *new_wav2mfcc(SP16 speech[], int speechlen)
{
  HTK_Param *param;
  Value para;                   /* parameters for Wav2MFCC */
  int framenum;
  int i;
  int len;

  /* set parameters */
  init_para(&para);

  if (ssload_filename && ssbuf == NULL) {
    /* load noise spectrum for spectral subtraction from file (once) */
    if ((ssbuf = new_SS_load_from_file(ssload_filename, &sslen)) == NULL) {
      j_error("Error: failed to read \"%s\"\n", ssload_filename);
    }
  }

  if (sscalc) {
    /* compute noise spectrum from head silence for each input */
    len = sscalc_len * smpFreq / 1000;
    if (len > speechlen) len = speechlen;
#ifdef SSDEBUG
    printf("[%d]\n", len);
#endif
    ssbuf = new_SS_calculate(speech, len, para, &sslen);
  }
#ifdef SSDEBUG
  {
    int i;
    for(i=0;i<sslen;i++) {
      printf("%d: %f\n", i, ssbuf[i]);
    }
  }
#endif
  
  /* calculate frame length from speech length, frame size and frame shift */
  framenum = (int)((speechlen - para.framesize) / para.frameshift) + 1;
  if (framenum < 1) {
    j_printerr("input too short (%d samples), ignored\n", speechlen);
    return NULL;
  }
  
  /* malloc new param */
  param = new_param();
  param->parvec = (VECT **)mymalloc(sizeof(VECT *) * framenum);
  for(i=0;i<framenum;i++) {
    param->parvec[i] = (VECT *)mymalloc(sizeof(VECT) * para.veclen);
  }

  /* make MFCC from speech data */
  Wav2MFCC(speech, param->parvec, para, speechlen, ssbuf, sslen);

  /* set miscellaneous parameters */
  param->header.samplenum = framenum;
  param->header.wshift = para.smp_period * para.frameshift;
  param->header.sampsize = para.veclen * sizeof(VECT); /* not compressed */
  param->header.samptype = F_MFCC;
  if (para.delta) param->header.samptype |= F_DELTA;
  if (para.acc) param->header.samptype |= F_ACCL;
  if (para.energy) param->header.samptype |= F_ENERGY;
  if (para.c0) param->header.samptype |= F_ZEROTH;
  if (para.absesup) param->header.samptype |= F_ENERGY_SUP;
  if (para.cmn) param->header.samptype |= F_CEPNORM;
  param->veclen = para.veclen;
  param->samplenum = framenum;

  return param;
}

---