void
segment_audio()
{
FILE *file;
int16 pcm_buf[BLOCKSIZE];
mfcc_t **cep_buf;
int16 voiced_buf = NULL;
int32 voiced_nsamps, out_frameidx, uttstart = 0;
char file_name[1024];
uint8 cur_vad_state, vad_state, writing;
int uttno, uttlen, sample_rate;
int32 nframes, nframes_tmp;
int16 frame_size, frame_shift, frame_rate;
size_t k;
sample_rate = (int) cmd_ln_float32_r(config, "-samprate");
frame_rate = cmd_ln_int32_r(config, "-frate");
frame_size =
(int32) (cmd_ln_float32_r(config, "-wlen") * sample_rate + 0.5);
frame_shift =
(int32) (sample_rate / cmd_ln_int32_r(config, "-frate") + 0.5);
nframes = (BLOCKSIZE - frame_size) / frame_shift;
cep_buf =
(mfcc_t **) ckd_calloc_2d(nframes, fe_get_output_size(fe),
sizeof(mfcc_t));
uttno = 0;
uttlen = 0;
cur_vad_state = 0;
voiced_nsamps = 0;
writing = 0;
file = NULL;
fe_start_stream(fe);
fe_start_utt(fe);
while ((k = read_audio(pcm_buf, BLOCKSIZE)) > 0) {
int16 const *pcm_buf_tmp;
pcm_buf_tmp = &pcm_buf[0];
while (k) {
nframes_tmp = nframes;
fe_process_frames_ext(fe, &pcm_buf_tmp, &k, cep_buf,
&nframes_tmp, voiced_buf,
&voiced_nsamps, &out_frameidx);
if (out_frameidx > 0) {
uttstart = out_frameidx;
}
vad_state = fe_get_vad_state(fe);
if (!cur_vad_state && vad_state) {
/* silence->speech transition, time to start new file */
uttno++;
if (!singlefile) {
sprintf(file_name, "%s%04d.raw", infile_path, uttno);
if ((file = fopen(file_name, "wb")) == NULL)
E_FATAL_SYSTEM("Failed to open '%s' for writing",
file_name);
} else {
sprintf(file_name, "%s.raw", infile_path);
if ((file = fopen(file_name, "ab")) == NULL)
E_FATAL_SYSTEM("Failed to open '%s' for writing",
file_name);
}
writing = 1;
}
if (writing && file && voiced_nsamps > 0) {
fwrite(voiced_buf, sizeof(int16), voiced_nsamps, file);
uttlen += voiced_nsamps;
}
if (cur_vad_state && !vad_state) {
/* speech -> silence transition, time to finish file */
fclose(file);
printf("Utterance %04d: file %s start %.1f sec length %d samples ( %.2f sec )\n",
uttno,
file_name,
((double) uttstart) / frame_rate,
uttlen,
((double) uttlen) / sample_rate);
fflush(stdout);
fe_end_utt(fe, cep_buf[0], &nframes_tmp);
writing = 0;
uttlen = 0;
voiced_nsamps = 0;
fe_start_utt(fe);
}
cur_vad_state = vad_state;
}
}
if (writing) {
fclose(file);
printf("Utterance %04d: file %s start %.1f sec length %d samples ( %.2f sec )\n",
uttno,
file_name,
((double) uttstart) / frame_rate,
uttlen,
((double) uttlen) / sample_rate);
fflush(stdout);
}
fe_end_utt(fe, cep_buf[0], &nframes);
ckd_free_2d(cep_buf);
}
uint8
ps_get_in_speech(ps_decoder_t *ps)
{
return fe_get_vad_state(ps->acmod->fe);
}
