int main(int argc, char *argv[])
{
struct CODEC2 *codec2;
FILE *fin;
FILE *fout;
short *buf;
unsigned char *bits;
int nsam, nbit, i, r;
for(i=0; i<10; i++) {
r = codec2_rand();
printf("[%d] r = %d\n", i, r);
}
if (argc != 3) {
printf("usage: %s InputRawSpeechFile OutputRawSpeechFile\n", argv[0]);
exit(1);
}
if ( (fin = fopen(argv[1],"rb")) == NULL ) {
fprintf(stderr, "Error opening input speech file: %s: %s.\n",
argv[1], strerror(errno));
exit(1);
}
if ( (fout = fopen(argv[2],"wb")) == NULL ) {
fprintf(stderr, "Error opening output speech file: %s: %s.\n",
argv[2], strerror(errno));
exit(1);
}
#ifdef DUMP
dump_on("c2demo");
#endif
/* Note only one set of Codec 2 states is required for an encoder
and decoder pair. */
codec2 = codec2_create(CODEC2_MODE_1300);
nsam = codec2_samples_per_frame(codec2);
buf = (short*)malloc(nsam*sizeof(short));
nbit = codec2_bits_per_frame(codec2);
bits = (unsigned char*)malloc(nbit*sizeof(char));
while(fread(buf, sizeof(short), nsam, fin) == (size_t)nsam) {
codec2_encode(codec2, bits, buf);
codec2_decode(codec2, buf, bits);
fwrite(buf, sizeof(short), nsam, fout);
}
free(buf);
free(bits);
codec2_destroy(codec2);
fclose(fin);
fclose(fout);
return 0;
}
void postfilter(
MODEL *model,
float *bg_est
)
{
int m, uv;
float e, thresh;
/* determine average energy across spectrum */
e = 1E-12;
for(m=1; m<=model->L; m++)
e += model->A[m]*model->A[m];
assert(e > 0.0);
e = 10.0*log10f(e/model->L);
/* If beneath threhold, update bg estimate. The idea
of the threshold is to prevent updating during high level
speech. */
if ((e < BG_THRESH) && !model->voiced)
*bg_est = *bg_est*(1.0 - BG_BETA) + e*BG_BETA;
/* now mess with phases during voiced frames to make any harmonics
less then our background estimate unvoiced.
*/
uv = 0;
thresh = powf(10.0, (*bg_est + BG_MARGIN)/20.0);
if (model->voiced)
for(m=1; m<=model->L; m++)
if (model->A[m] < thresh) {
model->phi[m] = TWO_PI*(float)codec2_rand()/CODEC2_RAND_MAX;
uv++;
}
#ifdef DUMP
dump_bg(e, *bg_est, 100.0*uv/model->L);
#endif
}
