/* phase vocoder by complex arithmetics with fixed hops.
* t_i - s_i = u_i - u_{i-1} = hop
* where s_i and t_i are the times for two analysis FFT
* and u_i is the time for the synthesis FFT at step i
* Reference: M.Puckette (1995)
* INPUT
* flag_lock : 0 == no phase lock is applied
* 1 == loose phase lock is applied
* rate : time-streching rate
* pitch_shift : in the unit of half-note
*/
void pv_complex (const char * file, const char * outfile,
double rate, double pitch_shift,
long len, long hop_syn,
int flag_window,
int flag_lock)
{
long hop_res = (long)((double)hop_syn * pow (2.0, - pitch_shift / 12.0));
long hop_ana = (long)((double)hop_res * rate);
long cur;
struct pv_complex * pv = pv_complex_init (len, hop_syn, flag_window);
SNDFILE * sf = NULL;
SF_INFO sfinfo;
SNDFILE * sfout = NULL;
SF_INFO sfout_info;
ao_device * ao = NULL;
pv->hop_res = hop_res;
pv->hop_ana = hop_ana;
/*pv->pitch_shift = pitch_shift; */
/* open file */
memset (&sfinfo, 0, sizeof (sfinfo));
sf = sf_open (file, SFM_READ, &sfinfo);
if (sf == NULL)
{
fprintf (stderr, "fail to open %s\n", file);
exit (1);
}
sndfile_print_info (&sfinfo);
pv_complex_set_input (pv, sf, &sfinfo);
if (outfile == NULL)
{
ao = ao_init_16_stereo (sfinfo.samplerate, 1 /* verbose */);
pv_complex_set_output_ao (pv, ao);
}
else
{
sfout = sndfile_open_for_write (&sfout_info,
outfile,
sfinfo.samplerate,
sfinfo.channels);
if (sfout == NULL)
{
fprintf (stderr, "fail to open file %s\n", outfile);
exit (1);
}
pv_complex_set_output_sf (pv, sfout, &sfout_info);
}
pv->flag_lock = flag_lock;
for (cur = 0; cur < (long)sfinfo.frames; cur += pv->hop_ana)
{
long len_play = pv_complex_play_step (pv, cur);
if (len_play < pv->hop_res)
{
break;
}
}
if (outfile == NULL)
{
ao_close (ao);
}
else
{
/* frames left in l_out[] and r_out[] */
sndfile_write (sfout, sfout_info,
pv->l_out, pv->r_out, len);
sf_write_sync (sfout);
sf_close (sfout);
}
pv_complex_free (pv);
sf_close (sf) ;
}
int main (int argc, char** argv)
{
extern int abs_flg; /* flag for absolute/relative cutoff */
extern double adj_pitch;
extern double pitch_shift;
extern int n_pitch;
char *file_midi = NULL;
char *file_wav = NULL;
char *file_patch = NULL;
int i;
// default value
double cut_ratio; // log10 of cutoff ratio for scale velocity
cut_ratio = -5.0;
double rel_cut_ratio; // log10 of cutoff ratio relative to average
rel_cut_ratio = 1.0; // this value is ignored when abs_flg == 1
long len = 2048;
int flag_window = 3; // hanning window
/* for 76 keys piano */
int notetop = 103; /* G8 */
int notelow = 28; /* E2 */
abs_flg = 1;
long hop = 0;
int show_help = 0;
int show_version = 0;
adj_pitch = 0.0;
/* to select peaks in a note */
int peak_threshold = 128; /* this means no peak search */
int flag_phase = 1; // use the phase correction
int psub_n = 0;
double psub_f = 0.0;
double oct_f = 0.0;
for (i = 1; i < argc; i++)
{
if ((strcmp (argv[i], "-input" ) == 0)
|| (strcmp (argv[i], "-i" ) == 0))
{
if ( i+1 < argc )
{
file_wav = (char *)malloc (sizeof (char)
* (strlen (argv[++i]) + 1));
CHECK_MALLOC (file_wav, "main");
strcpy (file_wav, argv[i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "-output" ) == 0)
|| (strcmp (argv[i], "-o" ) == 0))
{
if ( i+1 < argc )
{
file_midi = (char *)malloc (sizeof (char)
* (strlen (argv[++i]) + 1));
CHECK_MALLOC (file_midi, "main");
strcpy (file_midi, argv[i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--cutoff") == 0)
|| (strcmp (argv[i], "-c") == 0))
{
if ( i+1 < argc )
{
cut_ratio = atof (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--top") == 0)
|| (strcmp (argv[i], "-t") == 0))
{
if ( i+1 < argc )
{
notetop = atoi( argv[++i] );
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--bottom") == 0)
|| (strcmp (argv[i], "-b") == 0))
{
if ( i+1 < argc )
{
notelow = atoi (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--window") == 0)
|| (strcmp (argv[i], "-w") == 0))
{
if ( i+1 < argc )
{
flag_window = atoi (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if ( strcmp (argv[i], "-n") == 0)
{
if ( i+1 < argc )
{
len = atoi (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--shift") == 0)
|| (strcmp (argv[i], "-s") == 0))
{
if ( i+1 < argc )
{
hop = atoi (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--patch") == 0)
|| (strcmp (argv[i], "-p") == 0))
{
if ( i+1 < argc )
{
file_patch = argv[++i];
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--relative") == 0)
|| (strcmp (argv[i], "-r") == 0))
{
if ( i+1 < argc )
{
rel_cut_ratio = atof (argv[++i]);
abs_flg = 0;
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--peak") == 0)
|| (strcmp (argv[i], "-k") == 0))
{
if ( i+1 < argc )
{
peak_threshold = atoi (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--adjust") == 0)
|| (strcmp (argv[i], "-a") == 0))
{
if ( i+1 < argc )
{
adj_pitch = atof (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if ((strcmp (argv[i], "--help") == 0)
|| (strcmp (argv[i], "-h") == 0))
{
show_help = 1;
break;
}
else if (strcmp (argv[i], "-nophase") == 0)
{
flag_phase = 0;
}
else if (strcmp (argv[i], "-psub-n") == 0)
{
if ( i+1 < argc )
{
psub_n = atoi (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if (strcmp (argv[i], "-psub-f") == 0)
{
if ( i+1 < argc )
{
psub_f = atof (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if (strcmp (argv[i], "-oct") == 0)
{
if ( i+1 < argc )
{
oct_f = atof (argv[++i]);
}
else
{
show_help = 1;
break;
}
}
else if (strcmp (argv[i], "-v") == 0 ||
strcmp (argv[i], "--version") == 0)
{
show_version = 1;
}
else
{
show_help = 1;
}
}
if (show_help == 1)
{
print_usage (argv[0]);
exit (1);
}
else if (show_version == 1)
{
print_version ();
exit (1);
}
if (flag_window < 0 || flag_window > 6)
{
flag_window = 0;
}
if (hop == 0)
{
hop = len / 4;
}
if (psub_n == 0) psub_f = 0.0;
if (psub_f == 0.0) psub_n = 0;
struct WAON_notes *notes = WAON_notes_init();
CHECK_MALLOC (notes, "main");
char vel[128]; // velocity at the current step
int on_event[128]; // event index of struct WAON_notes.
for (i = 0; i < 128; i ++)
{
vel[i] = 0;
on_event[i] = -1;
}
// allocate buffers
double *left = (double *)malloc (sizeof (double) * len);
double *right = (double *)malloc (sizeof (double) * len);
CHECK_MALLOC (left, "main");
CHECK_MALLOC (right, "main");
double *x = NULL; /* wave data for FFT */
double *y = NULL; /* spectrum data for FFT */
#ifdef FFTW2
x = (double *)malloc (sizeof (double) * len);
y = (double *)malloc (sizeof (double) * len);
#else // FFTW3
x = (double *)fftw_malloc (sizeof (double) * len);
y = (double *)fftw_malloc (sizeof (double) * len);
#endif // FFTW2
CHECK_MALLOC (x, "main");
CHECK_MALLOC (y, "main");
/* power spectrum */
double *p = (double *)malloc (sizeof (double) * (len / 2 + 1));
CHECK_MALLOC (p, "main");
double *p0 = NULL;
double *dphi = NULL;
double *ph0 = NULL;
double *ph1 = NULL;
if (flag_phase != 0)
{
p0 = (double *)malloc (sizeof (double) * (len / 2 + 1));
CHECK_MALLOC (p0, "main");
dphi = (double *)malloc (sizeof (double) * (len / 2 + 1));
CHECK_MALLOC (dphi, "main");
ph0 = (double *)malloc (sizeof (double) * (len/2+1));
ph1 = (double *)malloc (sizeof (double) * (len/2+1));
CHECK_MALLOC (ph0, "main");
CHECK_MALLOC (ph1, "main");
}
double *pmidi = (double *)malloc (sizeof (double) * 128);
CHECK_MALLOC (pmidi, "main");
// MIDI output
if (file_midi == NULL)
{
file_midi = (char *)malloc (sizeof (char) * (strlen("output.mid") + 1));
CHECK_MALLOC (file_midi, "main");
strcpy (file_midi, "output.mid");
}
// open input wav file
if (file_wav == NULL)
{
file_wav = (char *) malloc (sizeof (char) * 2);
CHECK_MALLOC (file_wav, "main");
file_wav [0] = '-';
}
SF_INFO sfinfo;
SNDFILE *sf = sf_open (file_wav, SFM_READ, &sfinfo);
if (sf == NULL)
{
fprintf (stderr, "Can't open input file %s : %s\n",
file_wav, strerror (errno));
exit (1);
}
sndfile_print_info (&sfinfo);
// check stereo or mono
if (sfinfo.channels != 2 && sfinfo.channels != 1)
{
fprintf (stderr, "only mono and stereo inputs are supported.\n");
exit (1);
}
// time-period for FFT (inverse of smallest frequency)
double t0 = (double)len/(double)sfinfo.samplerate;
// weight of window function for FFT
double den = init_den (len, flag_window);
/* set range to analyse (search notes) */
/* -- after 't0' is calculated */
int i0 = (int)(mid2freq[notelow]*t0 - 0.5);
int i1 = (int)(mid2freq[notetop]*t0 - 0.5)+1;
if (i0 <= 0)
{
i0 = 1; // i0=0 means DC component (frequency = 0)
}
if (i1 >= (len/2))
{
i1 = len/2 - 1;
}
// init patch
init_patch (file_patch, len, flag_window);
/* ^^^ len could be given by option separately */
// initialization plan for FFTW
#ifdef FFTW2
rfftw_plan plan;
plan = rfftw_create_plan (len, FFTW_REAL_TO_COMPLEX, FFTW_ESTIMATE);
#else // FFTW3
fftw_plan plan;
plan = fftw_plan_r2r_1d (len, x, y, FFTW_R2HC, FFTW_ESTIMATE);
#endif
// for first step
if (hop != len)
{
if (sndfile_read (sf, sfinfo,
left + hop,
right + hop,
(len - hop))
!= (len - hop))
{
fprintf (stderr, "No Wav Data!\n");
exit(0);
}
}
/** main loop (icnt) **/
pitch_shift = 0.0;
n_pitch = 0;
int icnt; /* counter */
for (icnt=0; ; icnt++)
{
// shift
for (i = 0; i < len - hop; i ++)
{
if (sfinfo.channels == 2) // stereo
{
left [i] = left [i + hop];
right [i] = right [i + hop];
}
else // mono
{
left [i] = left [i + hop];
}
}
// read from wav
if (sndfile_read (sf, sfinfo,
left + (len - hop),
right + (len - hop),
hop)
!= hop)
{
fprintf (stderr, "WaoN : end of file.\n");
break;
}
// set double table x[] for FFT
for (i = 0; i < len; i ++)
{
if (sfinfo.channels == 2) // stereo
{
x [i] = 0.5 * (left [i] + right [i]);
}
else // mono
{
x [i] = left [i];
}
}
/**
* stage 1: calc power spectrum
*/
windowing (len, x, flag_window, 1.0, x);
/* FFTW library */
#ifdef FFTW2
rfftw_one (plan, x, y);
#else // FFTW3
fftw_execute (plan); // x[] -> y[]
#endif
if (flag_phase == 0)
{
// no phase-vocoder correction
HC_to_amp2 (len, y, den, p);
}
else
{
// with phase-vocoder correction
HC_to_polar2 (len, y, 0, den, p, ph1);
if (icnt == 0) // first step, so no ph0[] yet
{
for (i = 0; i < (len/2+1); ++i) // full span
{
// no correction
dphi[i] = 0.0;
// backup the phase for the next step
p0 [i] = p [i];
ph0 [i] = ph1 [i];
}
}
else // icnt > 0
{
// freq correction by phase difference
for (i = 0; i < (len/2+1); ++i) // full span
{
double twopi = 2.0 * M_PI;
//double dphi;
dphi[i] = ph1[i] - ph0[i]
- twopi * (double)i / (double)len * (double)hop;
for (; dphi[i] >= M_PI; dphi[i] -= twopi);
for (; dphi[i] < -M_PI; dphi[i] += twopi);
// frequency correction
// NOTE: freq is (i / len + dphi) * samplerate [Hz]
dphi[i] = dphi[i] / twopi / (double)hop;
// backup the phase for the next step
p0 [i] = p [i];
ph0 [i] = ph1 [i];
// then, average the power for the analysis
p[i] = 0.5 *(sqrt (p[i]) + sqrt (p0[i]));
p[i] = p[i] * p[i];
}
}
}
// drum-removal process
if (psub_n != 0)
{
power_subtract_ave (len, p, psub_n, psub_f);
}
// octave-removal process
if (oct_f != 0.0)
{
power_subtract_octave (len, p, oct_f);
}
/**
* stage 2: pickup notes
*/
/* new code
if (flag_phase == 0)
{
average_FFT_into_midi (len, (double)sfinfo.samplerate,
p, NULL,
pmidi);
}
else
{
average_FFT_into_midi (len, (double)sfinfo.samplerate,
p, dphi,
pmidi);
}
pickup_notes (pmidi,
cut_ratio, rel_cut_ratio,
notelow, notetop,
vel);
*/
/* old code */
if (flag_phase == 0)
{
// no phase-vocoder correction
note_intensity (p, NULL,
cut_ratio, rel_cut_ratio, i0, i1, t0, vel);
}
else
{
// with phase-vocoder correction
// make corrected frequency (i / len + dphi) * samplerate [Hz]
for (i = 0; i < (len/2+1); ++i) // full span
{
dphi[i] = ((double)i / (double)len + dphi[i])
* (double)sfinfo.samplerate;
}
note_intensity (p, dphi,
cut_ratio, rel_cut_ratio, i0, i1, t0, vel);
}
/**
* stage 3: check previous time for note-on/off
*/
WAON_notes_check (notes, icnt, vel, on_event,
8, 0, peak_threshold);
}
// clean notes
WAON_notes_regulate (notes);
WAON_notes_remove_shortnotes (notes, 1, 64);
WAON_notes_remove_shortnotes (notes, 2, 28);
WAON_notes_remove_octaves (notes);
/*
pitch_shift /= (double) n_pitch;
fprintf (stderr, "WaoN : difference of pitch = %f ( + %f )\n",
-(pitch_shift - 0.5),
adj_pitch);
*/
/* div is the divisions for one beat (quater-note).
* here we assume 120 BPM, that is, 1 beat is 0.5 sec.
* note: (hop / ft->rate) = duration for 1 step (sec) */
long div = (long)(0.5 * (double)sfinfo.samplerate / (double) hop);
fprintf (stderr, "division = %ld\n", div);
fprintf (stderr, "WaoN : # of events = %d\n", notes->n);
WAON_notes_output_midi (notes, div, file_midi);
#ifdef FFTW2
rfftw_destroy_plan (plan);
#else
fftw_destroy_plan (plan);
#endif /* FFTW2 */
WAON_notes_free (notes);
free (left);
free (right);
free (x);
free (y);
free (p);
if (p0 != NULL) free (p0);
if (dphi != NULL) free (dphi);
if (ph0 != NULL) free (ph0);
if (ph1 != NULL) free (ph1);
if (pmidi != NULL) free (pmidi);
if (file_wav != NULL) free (file_wav);
if (file_midi != NULL) free (file_midi);
sf_close (sf);
return 0;
}
