-
Notifications
You must be signed in to change notification settings - Fork 7
/
denoise.c
1193 lines (987 loc) · 35.3 KB
/
denoise.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*****************************************************************************
* Gnome Wave Cleaner Version 0.19
* Copyright (C) 2001 Jeffrey J. Welty
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*******************************************************************************/
/* denoise.c */
#include <errno.h>
#include <string.h>
#include <stdlib.h>
#include "gwc.h"
#include <math.h>
#include <float.h>
#include <unistd.h>
#include <sys/time.h>
//struct timeb start_time, middle_time, end_time ;
struct timeval start_time, middle_time, end_time ;
struct timezone tzp;
void start_timer(void)
{
gettimeofday(&start_time,&tzp);
}
void stop_timer(char *message)
{
gettimeofday(&end_time,&tzp) ;
{
double fstart = start_time.tv_sec + (double)start_time.tv_usec/1000000.0 ;
double fend = end_time.tv_sec + (double)end_time.tv_usec/1000000.0 ;
fprintf(stderr, "%s in %7.3lf real seconds\n", message, fend-fstart) ;
}
}
#ifdef OLD_TIMER
struct timeb start_time, middle_time, end_time ;
void start_timer(void)
{
ftime(&start_time) ;
}
void stop_timer(char *message)
{
ftime(&end_time) ;
{
double fstart = start_time.time + (double)start_time.millitm/1000.0 ;
double fend = end_time.time + (double)end_time.millitm/1000.0 ;
fprintf(stderr, "%s in %7.3lf real seconds\n", message, fend-fstart) ;
}
}
#endif
double bark_z[DENOISE_MAX_FFT] ;
double window_coef[DENOISE_MAX_FFT] ;
double sum_window_wgts[DENOISE_MAX_FFT] ;
double jg_upper[DENOISE_MAX_FFT][11] ;
double jg_lower[DENOISE_MAX_FFT][11] ;
/* double *two_way_probs[DENOISE_MAX_FFT] ; */
void compute_bark_z(int FFT_SIZE, int rate)
{
int k ;
/* compute the bark z value for this frequency bin */
for(k = 1 ; k <= FFT_SIZE/2 ; k++) {
double freq = (double)rate / 2.0 /(double)(FFT_SIZE/2)*(double)k ;
bark_z[k] = 7.0*log(freq/650.0 + sqrt(1 + (freq/650.0)*(freq/650.0))) ;
}
}
void compute_johnston_gain(int FFT_SIZE, double tonality_factor)
{
int k ;
for (k = 1; k <= FFT_SIZE/2 ; ++k) {
int j ;
for(j = k-1 ; j > 0 ; j--) {
double bark_diff = bark_z[k] - bark_z[j] ;
double johnston = 15.81 + 7.5*(bark_diff+.474) - 17.5*sqrt(1.0+(bark_diff+0.474)*(bark_diff+0.474)) ;
double johnston_masked = johnston - (tonality_factor*(14.5+bark_z[j])+5.5*(1.0 - tonality_factor)) ;
double gain = pow(10.0, johnston_masked/10.0) ;
jg_lower[k][k-j] = gain ;
if(k - j > 10) break ;
}
for(j = k ; j <= FFT_SIZE/2 ; j++) {
double bark_diff = bark_z[j] - bark_z[k] ;
double johnston = 15.81 + 7.5*(bark_diff+.474) - 17.5*sqrt(1.0+(bark_diff+0.474)*(bark_diff+0.474)) ;
double johnston_masked = johnston - (tonality_factor*(14.5+bark_z[j])+5.5*(1.0 - tonality_factor)) ;
double gain = pow(10.0, johnston_masked/10.0) ;
jg_upper[k][j-k] = gain ;
if(j - k > 10) break ;
}
}
}
int get_window_delta(struct denoise_prefs *pDnprefs)
{
if(pDnprefs->window_type == DENOISE_WINDOW_HANNING_OVERLAP_ADD) {
return pDnprefs->FFT_SIZE/2 ;
#ifdef DENOISE_TRY_ONE_SAMPLE
} else if(pDnprefs->window_type == DENOISE_WINDOW_ONE_SAMPLE) {
return 1 ;
#endif
} else {
if(pDnprefs->window_type == DENOISE_WINDOW_BLACKMAN)
return pDnprefs->FFT_SIZE/pDnprefs->smoothness ;
else
return 3*pDnprefs->FFT_SIZE/4 ;
}
}
void compute_sum_window_wgts(struct denoise_prefs *pDnprefs)
{
int delta = get_window_delta(pDnprefs) ;
int i, k ;
for(i = 0 ; i < pDnprefs->FFT_SIZE ; i++) {
sum_window_wgts[i] = 0.0 ;
for(k = i ; k < pDnprefs->FFT_SIZE+i ; k += delta) {
sum_window_wgts[i] += window_coef[k%pDnprefs->FFT_SIZE] ;
}
}
/* for(i = 0 ; i < pDnprefs->FFT_SIZE ; i++) { */
/* g_print("i:%d sww:%lf\n", i, sum_window_wgts[i]) ; */
/* } */
}
double gain_weiner(double Yk2, double Dk2)
{
double gain ;
double Xk2 = Yk2 - Dk2 ;
if(Yk2 > Dk2)
gain = (Xk2) / (Xk2+Dk2) ;
else
gain = 0.0 ;
return gain ;
}
double gain_power_subtraction(double Yk2, double Dk2)
{
double level = MAX(Yk2-Dk2, 0.0) ;
if(Yk2 > DBL_MIN)
return level/Yk2 ;
else
return 0.0 ;
}
double alpha_lorber(double snr)
{
double snr_db = 10.*log10(snr) ;
double alpha ;
if(snr_db > 20) return 1.0 ;
alpha = MIN((3.0 - 0.10*snr_db), 3.5) ;
return alpha ;
}
#define SLOW_EM
#ifdef SLOW_EM
double hypergeom(double theta)
{
double i0(double), i1(double) ;
if(theta < 7.389056)
return exp(-theta/2.0)*(1.0+theta*i0(theta/2.0)+theta*i1(theta/2.0)) ;
else
return exp(0.09379 + 0.50447*log(theta)) ;
}
double gain_em(double Rprio, double Rpost, double alpha)
{
/* Ephraim-Malah classic noise suppression, from 1984 paper */
double gain = 0.886226925*sqrt(1.0/(1.0+Rpost)*(Rprio/(1.0+Rprio))) ;
gain *= hypergeom((1.0+Rpost)*(Rprio/(1.0+Rprio))) ;
return gain ;
}
#else
double gain_em(double Rprio, double Rpost, double alpha)
{
/* Ephraim-Malah noise suppression, from Godsill and Wolfe 2001 paper */
double r = MAX(Rprio/(1.0+Rprio),DBL_MIN) ;
double V = Rprio/(1.+Rprio)*Rpost ;
return sqrt( r * (1.0+V)/Rpost ) ;
}
#endif
double blackman(int k, int N)
{
double p = ((double)(k))/(double)(N-1) ;
return 0.42-0.5*cos(2.0*M_PI*p) + 0.08*cos(4.0*M_PI*p) ;
}
double hanning(int k, int N)
{
double p = ((double)(k))/(double)(N-1) ;
return 0.5 - 0.5 * cos(2.0*M_PI*p) ;
}
double blackman_hybrid(int k, int n_flat, int N)
{
if(k >= (N-n_flat)/2 && k <= n_flat+(N-n_flat)/2-1) {
return 1.0 ;
} else {
double p ;
if(k >= n_flat+(N-n_flat)/2-1) k -= n_flat ;
p = (double)(k)/(double)(N-n_flat-1) ;
return 0.42-0.5*cos(2.0*M_PI*p) + 0.08*cos(4.0*M_PI*p) ;
}
}
double welty_alpha(double w, double x)
{
double alpha = ( log(acos(-2.0*w+1)) - log(M_PI) ) / log(1.0 - x) ;
/* d_print("Welty alpha=%g\n", alpha) ; */
return alpha ;
}
/* double welty(int k, int N, double alpha) */
/* { */
/* double n2 = (double)N/2.0 ; */
/* double x = fabs(((double)k - n2) / (n2)) ; */
/* double tx = pow(1.0-x, alpha)*M_PI ; */
/* double w = -( cos(tx)-1.0 )/2.0 ; */
/* d_print("Welty x=%g, w=%g, k=%d N=%d\n", x, w, k, N) ; */
/* } */
double fft_window(int k, int N, int window_type)
{
if(window_type == DENOISE_WINDOW_BLACKMAN) {
return blackman(k, N) ;
} else if(window_type == DENOISE_WINDOW_BLACKMAN_HYBRID) {
return blackman_hybrid(k, N-N/4, N) ;
#ifdef DENOISE_TRY_ONE_SAMPLE
} else if(window_type == DENOISE_WINDOW_ONE_SAMPLE) {
return hanning(k, N) ;
#endif
} else if(window_type == DENOISE_WINDOW_HANNING_OVERLAP_ADD) {
return hanning(k, N) ;
}
return 0.0 ;
}
double db2w(double db)
{
return pow(10,db/10) ;
}
double ceramic_mic_response(double f)
{
double db = 0.0 ;
if(f < 4000.0) {
db = 0.0 ;
} else if(f < 7000.0) {
db = (f-4000.0)/7000.0*2.0 ;
} else if(f < 20000.0) {
db = 2.0 - ((f-7000.0)/13000.0)*12.0 ;
} else {
db = -10.0 ;
}
return db2w(db) ;
}
void cdivide(double *a, double *b, double c, double d)
{
double denom = (c*c + d*d) ;
double anew, bnew ;
if(denom < 1.e-60) {
*a = 0.0 ;
*b = 0.0 ;
return ;
}
anew = (*a*c + *b*d) / denom ;
bnew = (*b*c - *a*d) / denom ;
}
#define bin2freq(r,s,k) ((double)r / 2.0 /(double)(s/2)*(double)k)
int prev_sample[2] ;
static fftw_real windowed[DENOISE_MAX_FFT] ;
static fftw_real out[DENOISE_MAX_FFT] ;
#ifdef HAVE_FFTW3
static void fft_remove_noise(fftw_real sample[], fftw_real noise_min2[], fftw_real noise_max2[], fftw_real noise_avg2[],
FFTW(plan) *pFor, FFTW(plan) *pBak,
#else /* HAVE_FFTW3 */
void fft_remove_noise(fftw_real sample[], fftw_real noise_min2[], fftw_real noise_max2[], fftw_real noise_avg2[],
rfftw_plan *pFor, rfftw_plan *pBak,
#endif /* HAVE_FFTW3 */
struct denoise_prefs *pPrefs, int ch)
{
int k ;
fftw_real noise2[DENOISE_MAX_FFT/2+1] ;
fftw_real noise[DENOISE_MAX_FFT/2+1] ;
fftw_real Y2[DENOISE_MAX_FFT/2+1] ;
fftw_real Y[DENOISE_MAX_FFT/2+1] ;
fftw_real masked[DENOISE_MAX_FFT/2+1] ;
fftw_real gain_k[DENOISE_MAX_FFT] ;
static fftw_real bsig_prev[2][DENOISE_MAX_FFT],bY2_prev[2][DENOISE_MAX_FFT/2+1],bgain_prev[2][DENOISE_MAX_FFT/2+1] ;
fftw_real *sig_prev,*Y2_prev,*gain_prev ;
static int debug_frame = 1 ;
double SFM, tonality_factor ;
sig_prev = bsig_prev[ch] ;
Y2_prev = bY2_prev[ch] ;
gain_prev = bgain_prev[ch] ;
if(0 || pPrefs->window_type == DENOISE_WINDOW_HANNING_OVERLAP_ADD) {
/* with overlap-add pre-window the sample */
for(k = 0 ; k < pPrefs->FFT_SIZE ; k++) {
windowed[k] = sample[k]*window_coef[k] ;
//if(debug_frame == 3) g_print("k:%d window_coef:%lf windowed:%lf\n", k, window_coef[k], windowed[k]) ;
}
} else {
/* with other methods don't window the sample for FFT,
but window the result back into the original
sample the FFT noise-removal process */
for(k = 0 ; k < pPrefs->FFT_SIZE ; k++) {
windowed[k] = sample[k] ;
}
}
#ifdef HAVE_FFTW3
FFTW(execute)(*pFor);
#else /* HAVE_FFTW3 */
rfftw_one(*pFor, windowed, out);
#endif /* HAVE_FFTW3 */
{
double sum_log_p = 0.0 ;
double sum_p = 0.0 ;
double kinv = 1./(double)(pPrefs->FFT_SIZE/2.0) ;
for (k = 1; k <= pPrefs->FFT_SIZE/2 ; ++k) {
noise2[k] = noise_max2[k] ;
noise2[k] = noise_min2[k] + 0.5*(noise_max2[k] - noise_min2[k]) ;
noise2[k] = noise_avg2[k] ;
noise[k] = sqrt(noise2[k]) ;
if(k < pPrefs->FFT_SIZE/2) {
Y2[k] = out[k]*out[k] + out[pPrefs->FFT_SIZE-k]*out[pPrefs->FFT_SIZE-k] ;
Y[k] = sqrt(Y2[k]) ;
} else {
Y2[k] = out[k]*out[k] ;
Y[k] = out[k] ;
}
sum_log_p += log10(Y2[k]) ;
sum_p += Y2[k] ;
}
SFM = 10.0*( kinv*sum_log_p - log10(sum_p*kinv) ) ;
tonality_factor = MIN(SFM/-60.0, 1) ;
}
if(pPrefs->noise_suppression_method == DENOISE_LORBER) tonality_factor = 0.0 ;
/* g_print("SFM:%f tonality:%lf\n", SFM, tonality_factor) ; */
if(pPrefs->noise_suppression_method == DENOISE_LORBER) {
for (k = 1; k <= pPrefs->FFT_SIZE/2 ; ++k) {
double sum = 0 ;
double sum_wgts = 0 ;
int j ;
int j1 = MAX(k-10,1) ;
int j2 = MIN(k+10,pPrefs->FFT_SIZE/2) ;
for(j = j1 ; j <= j2 ; j++) {
double d = ABS(j-k)+1.0 ;
double wgt = 1./sqrt(d) ;
sum += Y2[j]*wgt ;
sum_wgts += wgt ;
}
masked[k] = sum / (sum_wgts+1.e-300) ;
}
}
/* if(pPrefs->noise_suppression_method == DENOISE_LORBER || pPrefs->noise_suppression_method == DENOISE_WOLFE_GODSILL) { */
if(pPrefs->noise_suppression_method == DENOISE_WOLFE_GODSILL) {
for (k = 1; k <= pPrefs->FFT_SIZE/2 ; ++k) {
int j ;
masked[k] = 0.0 ;
for(j = k-1 ; j > 0 ; j--) {
#ifdef OLD_N_SLOW
double bark_diff = bark_z[k] - bark_z[j] ;
double johnston = 15.81 + 7.5*(bark_diff+.474) - 17.5*sqrt(1.0+(bark_diff+0.474)*(bark_diff+0.474)) ;
double johnston_masked = johnston - (tonality_factor*(14.5+bark_z[j])+5.5*(1.0 - tonality_factor)) ;
double gain = pow(10.0, johnston_masked/10.0) ;
if(gain < 1.e-2) break ;
#else
double gain = jg_lower[k][k-j] ;
#endif
if(k - j > 10) break ;
masked[k] += MAX((Y2[j]-noise2[j]),0.0)*gain ;
}
for(j = k ; j <= pPrefs->FFT_SIZE/2 ; j++) {
#ifdef OLD_N_SLOW
double bark_diff = bark_z[j] - bark_z[k] ;
double johnston = 15.81 + 7.5*(bark_diff+.474) - 17.5*sqrt(1.0+(bark_diff+0.474)*(bark_diff+0.474)) ;
double johnston_masked = johnston - (tonality_factor*(14.5+bark_z[j])+5.5*(1.0 - tonality_factor)) ;
double gain = pow(10.0, johnston_masked/10.0) ;
#else
double gain = jg_upper[k][j-k] ;
#endif
if(gain < 1.e-2) break ;
if(j - k > 10) break ;
masked[k] += MAX((Y2[j]-noise2[j]),0.0)*gain ;
}
/* if(debug_frame == 3) g_print("k:%d Y:%lf masked:%lf bark:%lf\n", k, Y[k], masked[k], bark_z[k]) ; */
}
}
#ifdef TEST
if(pPrefs->noise_suppression_method == DENOISE_AUDACITY) {
for(i=0; i<=len/2; i++)
plog[i] = log(power[i]);
int half = len/2;
for(i=0; i<=half; i++) {
float smooth;
if (plog[i] < noiseGate[i] + (level/2.0))
smooth = 0.0;
else
smooth = 1.0;
smoothing[i] = smooth * 0.5 + smoothing[i] * 0.5;
}
/* try to eliminate tinkle bells */
for(i=2; i<=half-2; i++) {
if (smoothing[i]>=0.5 &&
smoothing[i]<=0.55 &&
smoothing[i-1]<0.1 &&
smoothing[i-2]<0.1 &&
smoothing[i+1]<0.1 &&
smoothing[i+2]<0.1)
smoothing[i] = 0.0;
}
outr[0] *= smoothing[0];
outi[0] *= smoothing[0];
outr[half] *= smoothing[half];
outi[half] *= smoothing[half];
for(i=1; i<half; i++) {
int j = len - i;
float smooth = smoothing[i];
outr[i] *= smooth;
outi[i] *= smooth;
outr[j] *= smooth;
outi[j] *= smooth;
}
#else
if(0) {
#endif
} else {
if(pPrefs->noise_suppression_method == DENOISE_EXPERIMENTAL) {
/* this whole section is starting to play around with audio *restoration*, i.e. trying
* to guess what was missing, and adding it back in. Think of very old recordings
* with that have a damped high frequency response */
if(0) {
static int first=1 ;
/* try simple deconvolution */
for (k = 0 ; k <= pPrefs->FFT_SIZE/2 ; ++k) {
/* Gk needs to look like the original response filter */
double Gk = hanning(k,pPrefs->FFT_SIZE/2) ;
double freq = bin2freq(pPrefs->rate,pPrefs->FFT_SIZE,k) ;
Gk = ceramic_mic_response(bin2freq(pPrefs->rate,pPrefs->FFT_SIZE,k)) ;
Gk = Gk * Gk ;
if( !(k % 100) && first ) printf("k:%d freq:%0.1f Gk:%f\n", k, freq, Gk) ;
if(k < pPrefs->FFT_SIZE/2) {
/* cdivide(&out[k],&out[pPrefs->FFT_SIZE-k],Gk,0) ; */
out[pPrefs->FFT_SIZE-k] /= Gk ;
out[k] /= Gk ;
} else {
out[k] /= Gk ;
}
out[k] *= Gk ;
}
first=0 ;
} else {
// Step 1. Compute the linear regression coefs(a,b) for the function ln(power2) = a*k + b ;
double sumx = 0.0 ;
double sumy = 0.0 ;
double sumx2 = 0.0 ;
double sumxy = 0.0 ;
double n = 0 ;
double a, b ;
int k_cut_off = 2*pPrefs->FFT_SIZE/8 ;
fftw_real *r = gain_k ;
for (k = 1 ; k <= pPrefs->FFT_SIZE/2 ; ++k) {
r[k] = 1.0 ;
}
for (k = 1 ; k < k_cut_off ; ++k) {
double y = log(Y2[k]) ;
sumx += (double)k ;
sumy += y ;
sumxy += (double)k*y ;
sumx2 += (double)k*(double)k ;
n++ ;
}
a = (n*(sumxy)-sumx*sumy) / (n*sumx2 - sumx*sumx) ;
b = (sumy - a*sumx) / n ;
printf("a,b %f %f\n", a, b) ;
// Step 2. Compute r[k], amount of energy present for each frequency bin relative to predicted
for (k = 1 ; k < k_cut_off ; ++k) {
r[k] = log(Y2[k]) / (a*(double)k+b) ;
}
#define MAX_HARMONIC 3
// Step 3. Compute gain in each bin >= k_cutoff ;
double harmonic_factor[6] = {1.0,1.0,1.0,1.0,1.0,1.0/8.0} ;
int harmonic ;
for(harmonic = 1 ; harmonic < MAX_HARMONIC+1 ; harmonic++) {
double harmonic_fraction = 1.0 - ((double)harmonic / (double)(harmonic+1)) ;
for (k = k_cut_off ; k <= pPrefs->FFT_SIZE/2 ; ++k) {
/* compute j, the index of the fundamental freq of which k is the harmonic */
int j = (int)((double)k * harmonic_fraction + 0.5) ;
double ln_p2_hat = r[j]*(a * (double)k + b) ;
double p2_hat = harmonic_factor[harmonic]*exp(ln_p2_hat) ;
double Gk = sqrt((p2_hat+Y2[k]) / Y2[k]) ;
if (k == pPrefs->FFT_SIZE/2) printf("harmonic, hf, j, p2_hat, Gk[maxfft], ln_p2_hat %d %f %d %f %f %f\n",
harmonic, harmonic_fraction, j, p2_hat, Gk, ln_p2_hat) ;
out[k] *= Gk ;
if(k < pPrefs->FFT_SIZE/2) out[pPrefs->FFT_SIZE-k] *= Gk ;
}
}
}
} else {
for (k = 1; k <= pPrefs->FFT_SIZE/2 ; ++k) {
if(noise2[k] > DBL_MIN) {
double gain, Fk, Gk ;
if(pPrefs->noise_suppression_method == DENOISE_EM) {
double Rpost = MAX(Y2[k]/noise2[k]-1.0, 0.0) ;
double alpha = pPrefs->dn_gamma ;
double Rprio ;
if(prev_sample[ch] == 1)
Rprio = (1.0-alpha)*Rpost+alpha*gain_prev[k]*gain_prev[k]*Y2_prev[k]/noise2[k] ;
else
Rprio = Rpost ;
gain = gain_em(Rprio, Rpost, alpha) ;
/* g_print("Rpost:%lg Rprio:%lg gain:%lg gain_prev:%lg y2_prev:%lg\n", Rpost, Rprio, gain, gain_prev[k], Y2_prev[k]) ; */
gain_prev[k] = gain ;
Y2_prev[k] = Y2[k] ;
} else if(pPrefs->noise_suppression_method == DENOISE_LORBER) {
double SNRlocal = MAX(masked[k]/noise2[k]-1.0, 0.0) ;
double SNRfilt, SNRprio ;
double alpha ;
if(prev_sample[ch] == 1) {
double eta = (1.0 + pPrefs->dn_gamma) / 2.0 ; /* eta seems to be better closer to 1.0 than gamma */
SNRfilt = (1.-pPrefs->dn_gamma)*SNRlocal + pPrefs->dn_gamma*(sig_prev[k]/noise2[k]) ;
SNRprio = SNRfilt ; /* note, could use another parameter, like pPrefs->dn_gamma, here to compute SNRprio */
SNRprio = (1.-eta)*SNRlocal + eta*(sig_prev[k]/noise2[k]) ;
}else {
SNRfilt = SNRlocal ;
SNRprio = SNRfilt ;
}
alpha = alpha_lorber(SNRprio) ;
gain = 1.0 - alpha/(SNRfilt+1) ;
sig_prev[k] = MAX(Y2[k]*gain,0.0) ;
} else if(pPrefs->noise_suppression_method == DENOISE_WOLFE_GODSILL) {
double Rpost = MAX(Y2[k]/noise2[k]-1.0, 0.0) ;
double alpha = pPrefs->dn_gamma ;
double Rprio ;
if(prev_sample[ch] == 1)
Rprio = (1.0-alpha)*Rpost+alpha*gain_prev[k]*gain_prev[k]*Y2_prev[k]/noise2[k] ;
else
Rprio = Rpost ;
Y2_prev[k] = Y2[k] ;
if(Y2[k] > masked[k]) {
gain = MAX(masked[k]/Y2[k], Rprio/(Rprio+1.0)) ;
} else {
gain = 1.0 ;
}
gain_prev[k] = gain ;
} else if(pPrefs->noise_suppression_method == DENOISE_WEINER)
gain = gain_weiner(Y2[k], noise2[k]) ;
else
gain = gain_power_subtraction(Y2[k], noise2[k]) ;
Fk = pPrefs->amount*(1.0-gain) ;
if(Fk < 0.0) Fk = 0.0 ;
if(Fk > 1.0) Fk = 1.0 ;
Gk = 1.0 - Fk ;
out[k] *= Gk ;
if(k < pPrefs->FFT_SIZE/2) out[pPrefs->FFT_SIZE-k] *= Gk ;
gain_k[k] = Gk ;
}
}
}
}
/* the inverse fft */
#ifdef HAVE_FFTW3
FFTW(execute)(*pBak);
#else /* HAVE_FFTW3 */
rfftw_one(*pBak, out, windowed);
#endif /* !HAVE_FFTW3 */
for(k = 0 ; k < pPrefs->FFT_SIZE ; k++)
windowed[k] /= (double)(pPrefs->FFT_SIZE) ;
if(0|| pPrefs->window_type == DENOISE_WINDOW_HANNING_OVERLAP_ADD) {
/** HANNING - OVERLAP - ADD */
/* make sure the tails of the sample approach zero magnitude */
double offset = windowed[0] ;
double hs1 = pPrefs->FFT_SIZE/2 - 1 ;
for(k = 0 ; k < pPrefs->FFT_SIZE/2 ; k++) {
double p = (hs1-(double)k)/hs1 ;
sample[k] = windowed[k] - offset*p ;
}
offset = windowed[pPrefs->FFT_SIZE-1] ;
for(k = pPrefs->FFT_SIZE/2 ; k < pPrefs->FFT_SIZE ; k++) {
double p = ((double)k-hs1)/hs1 ;
sample[k] = windowed[k] - offset*p ;
}
} else {
/* merge results back into sample data based on window function */
for(k = 0 ; k < pPrefs->FFT_SIZE ; k++) {
double w = window_coef[k] ;
sample[k] = (1.0-w)*sample[k] + w*windowed[k] ;
}
}
prev_sample[ch] = 1 ;
debug_frame++ ;
}
int denoise_normalize = 0 ;
#define NO_DEBUG
#ifdef DEBUG
void print_denoise(char *header, struct denoise_prefs *pDnprefs)
{
int k ;
g_print("******** %s ************\n", header) ;
g_print("FFT_SIZE:%d\n", pDnprefs->FFT_SIZE) ;
g_print("n_noise_samples:%d\n", pDnprefs->n_noise_samples) ;
g_print("amount:%lf\n", pDnprefs->amount) ;
g_print("smoothness:%d\n", pDnprefs->smoothness) ;
g_print("window_type:") ;
switch(pDnprefs->window_type) {
case DENOISE_WINDOW_BLACKMAN : g_print("Blackman\n") ; break ;
case DENOISE_WINDOW_BLACKMAN_HYBRID : g_print("Blackman-hybrid\n") ; break ;
case DENOISE_WINDOW_HANNING_OVERLAP_ADD : g_print("Hanning-overlap-add\n") ; break ;
#ifdef DENOISE_TRY_ONE_SAMPLE
case DENOISE_WINDOW_ONE_SAMPLE : g_print("One Sample\n") ; break ;
#endif
default : g_print("!!!!!!!!! UNKNOWN !!!!!!!!!!!!\n") ; break ;
}
g_print("Suppression method:") ;
switch(pDnprefs->noise_suppression_method) {
case DENOISE_LORBER : g_print("Lorber-Hoeldrich\n") ; break ;
case DENOISE_WEINER : g_print("Weiner\n") ; break ;
case DENOISE_EM : g_print("Ephram\n") ; break ;
case DENOISE_WOLFE_GODSILL : g_print("Wolfe-Godsill\n") ; break ;
default : g_print("Spectral Subtraction\n") ; break ;
}
for(k = 0 ; k < pDnprefs->FFT_SIZE ; k++) {
g_print("k:%d wc:%lg\n", k, window_coef[k]) ;
}
}
#else
void print_denoise(char *header, struct denoise_prefs *pDnprefs) {}
#endif
void get_noise_sample(struct sound_prefs *pPrefs, struct denoise_prefs *pDnprefs,
long noise_start, long noise_end,
fftw_real *left_noise_min, fftw_real *left_noise_max, fftw_real *left_noise_avg,
fftw_real *right_noise_min, fftw_real *right_noise_max, fftw_real *right_noise_avg) ;
int denoise(struct sound_prefs *pPrefs, struct denoise_prefs *pDnprefs, long noise_start, long noise_end,
long first_sample, long last_sample, int channel_mask) {
long current ;
int k ;
fftw_real left[DENOISE_MAX_FFT], right[DENOISE_MAX_FFT] ;
fftw_real left_noise_max[DENOISE_MAX_FFT], right_noise_max[DENOISE_MAX_FFT], left_noise_avg[DENOISE_MAX_FFT] ;
fftw_real left_noise_min[DENOISE_MAX_FFT], right_noise_min[DENOISE_MAX_FFT], right_noise_avg[DENOISE_MAX_FFT] ;
fftw_real tmp[DENOISE_MAX_FFT] ;
fftw_real left_prev_frame[DENOISE_MAX_FFT] ;
fftw_real right_prev_frame[DENOISE_MAX_FFT] ;
#ifdef HAVE_FFTW3
FFTW(plan) pForLeft, pForRight ;
FFTW(plan) pFor, pBak ;
#else /* HAVE_FFTW3 */
rfftw_plan pFor, pBak ;
#endif /* HAVE_FFTW3 */
int framenum = 0 ;
double alpha ;
double s_amount ; /* amount, reduced to account for oversampling
due to smoothness */
start_timer() ;
current = first_sample ;
push_status_text("Denoising audio") ;
update_progress_bar(0.0,PROGRESS_UPDATE_INTERVAL,TRUE) ;
#ifdef HAVE_FFTW3
pFor =
FFTW(plan_r2r_1d)(pDnprefs->FFT_SIZE, windowed, out, FFTW_R2HC, FFTW_ESTIMATE);
pBak =
FFTW(plan_r2r_1d)(pDnprefs->FFT_SIZE, out, windowed, FFTW_HC2R, FFTW_ESTIMATE);
#endif /* HAVE_FFTW3 */
#ifdef HAVE_FFTW3
pForLeft =
FFTW(plan_r2r_1d)(pDnprefs->FFT_SIZE, left, tmp, FFTW_R2HC, FFTW_ESTIMATE);
pForRight =
FFTW(plan_r2r_1d)(pDnprefs->FFT_SIZE, right, tmp, FFTW_R2HC, FFTW_ESTIMATE);
#else /* HAVE_FFTW3 */
pFor = rfftw_create_plan(pDnprefs->FFT_SIZE, FFTW_REAL_TO_COMPLEX, FFTW_ESTIMATE);
pBak = rfftw_create_plan(pDnprefs->FFT_SIZE, FFTW_COMPLEX_TO_REAL, FFTW_ESTIMATE);
#endif /* HAVE_FFTW3 */
alpha = welty_alpha(0.5, 1.0/(double)pDnprefs->smoothness) ;
alpha = 1.0 ;
for(k = 0 ; k < pDnprefs->FFT_SIZE ; k++) {
window_coef[k] = fft_window(k,pDnprefs->FFT_SIZE, pDnprefs->window_type) ;
d_print("k:%d wc:%lg\n", k, window_coef[k]) ;
left_prev_frame[k] = 0.0 ;
left[k] = 0.0 ;
right_prev_frame[k] = 0.0 ;
right[k] = 0.0 ;
/* two_way_probs[k] = (double *)calloc(pDnprefs->FFT_SIZE,sizeof(double)) ; */
/* if(two_way_probs[k] == NULL) { */
/* fprintf(stderr, "Failed to malloc two_way_probs, %d of %d\n", k+1, pDnprefs->FFT_SIZE) ; */
/* exit(1) ; */
/* } */
}
audio_normalize(denoise_normalize) ;
get_noise_sample(pPrefs, pDnprefs, noise_start, noise_end,
left_noise_min, left_noise_max, left_noise_avg,
right_noise_min, right_noise_max, right_noise_avg) ;
pDnprefs->rate = pPrefs->rate ;
audio_normalize(denoise_normalize) ;
/* if(smoothness <= 4 || window_type == DENOISE_WINDOW_BLACKMAN_HYBRID) */
s_amount = pDnprefs->amount ;
/* else */
/* s_amount = amount/(double)(smoothness-3) ; */
prev_sample[0] = 0 ;
prev_sample[1] = 0 ;
if(pDnprefs->amount > DBL_MIN) {
while(current <= last_sample-pDnprefs->FFT_SIZE) {
long n = pDnprefs->FFT_SIZE ;
long tmplast = current + n - 1 ;
gfloat p = (gfloat)(current-first_sample)/(last_sample-first_sample) ;
n = read_fft_real_wavefile_data(left, right, current, current+pDnprefs->FFT_SIZE-1) ;
if(n < pDnprefs->FFT_SIZE) break ; /* hit the end of all data? */
#ifdef MAC_OS_X
// This usleep fixes a segfault on OS X. Rob Frohne
usleep(2);
#endif
update_progress_bar(p,PROGRESS_UPDATE_INTERVAL,FALSE) ;
/*
*/
if(framenum == 0) {
for(k = 0 ; k < pDnprefs->FFT_SIZE ; k++) {
if(k < pDnprefs->FFT_SIZE/2) {
d_print("OA sum %d:%lg\n", k, window_coef[k]+window_coef[k+pDnprefs->FFT_SIZE/2]) ;
window_coef[k] = 1.0 ;
}
}
compute_sum_window_wgts(pDnprefs) ;
framenum = 1 ;
} else if(framenum == 1) {
for(k = 0 ; k < pDnprefs->FFT_SIZE ; k++) {
window_coef[k] = fft_window(k,pDnprefs->FFT_SIZE, pDnprefs->window_type) ;
}
compute_sum_window_wgts(pDnprefs) ;
framenum = 2 ;
}
if(channel_mask & 0x01)
fft_remove_noise(left, left_noise_min, left_noise_max, left_noise_avg,
&pFor, &pBak, pDnprefs,0) ;
if(channel_mask & 0x02)
fft_remove_noise(right, right_noise_min, right_noise_max, right_noise_avg,
&pFor, &pBak, pDnprefs,1) ;
if(pDnprefs->window_type == DENOISE_WINDOW_HANNING_OVERLAP_ADD) {
for(k = 0 ; k < pDnprefs->FFT_SIZE/2 ; k++) {
if(channel_mask & 0x01) {
/* add in the last half of the previous output frame */
left[k] += left_prev_frame[k+pDnprefs->FFT_SIZE/2] ;
/* save the last half of this output frame */
left_prev_frame[k+pDnprefs->FFT_SIZE/2] = left[k+pDnprefs->FFT_SIZE/2] ;
}
if(channel_mask & 0x02) {
/* add in the last half of the previous output frame */
right[k] += right_prev_frame[k+pDnprefs->FFT_SIZE/2] ;
/* save the last half of this output frame */
right_prev_frame[k+pDnprefs->FFT_SIZE/2] = right[k+pDnprefs->FFT_SIZE/2] ;
}
}
write_fft_real_wavefile_data(left, right, current, current+pDnprefs->FFT_SIZE/2-1) ;
#ifdef DENOISE_TRY_ONE_SAMPLE
} else if(pDnprefs->window_type == DENOISE_WINDOW_ONE_SAMPLE) {
if(current == first_sample) {
write_fft_real_wavefile_data(left, right, current, current+pDnprefs->FFT_SIZE/2-1) ;
} else {
write_fft_real_wavefile_data(left, right, current, current) ;
}
#endif
} else {
write_fft_real_wavefile_data(left, right, current, MIN(tmplast, last_sample)) ;
}
current += get_window_delta(pDnprefs) ;
}
}
#ifdef HAVE_FFTW3
FFTW(destroy_plan)(pForLeft);
FFTW(destroy_plan)(pForRight);
FFTW(destroy_plan)(pBak);
FFTW(destroy_plan)(pFor);
#else /* HAVE_FFTW3 */
rfftw_destroy_plan(pFor);
rfftw_destroy_plan(pBak);
#endif /* HAVE_FFTW3 */
for(k = 0 ; k < pDnprefs->FFT_SIZE ; k++) {
/* free(two_way_probs[k]) ; */
}
audio_normalize(1) ;
update_progress_bar(0.0,PROGRESS_UPDATE_INTERVAL,TRUE) ;
pop_status_text() ;
main_redraw(FALSE, TRUE) ;
stop_timer("DENOISE") ;
return 0 ;
}
/* AJH: print_noise_sample is currently used by the DSP filter dialog's "show response" button; I suspect noise.dat is not an accurate name for the output */
/* but I'm pretty sure we should be graphing the output instead, anyway */
/* Also, there's currently no need for this function to return an int rather than void */
int print_noise_sample(struct sound_prefs *pPrefs, struct denoise_prefs *pDnprefs, long noise_start, long noise_end)
{
int k ;
FILE *fp ;
fftw_real left_noise_max[DENOISE_MAX_FFT], right_noise_max[DENOISE_MAX_FFT], left_noise_avg[DENOISE_MAX_FFT] ;
fftw_real left_noise_min[DENOISE_MAX_FFT], right_noise_min[DENOISE_MAX_FFT], right_noise_avg[DENOISE_MAX_FFT] ;
extern int MAXSAMPLEVALUE ;
double max = MAXSAMPLEVALUE * MAXSAMPLEVALUE ;
get_noise_sample(pPrefs, pDnprefs, noise_start, noise_end,
left_noise_min, left_noise_max, left_noise_avg,
right_noise_min, right_noise_max, right_noise_avg) ;
fp = fopen("noise.dat", "w") ;
if (fp == NULL) {
warning(g_strconcat("Cannot write to ", g_get_current_dir(), "/noise.dat: ", strerror(errno), NULL)); //this check prevents a segfault if we can't write to the file
return 1 ;
}
fprintf(stderr, "FFT_SIZE in print_noise_sample is %d\n", pDnprefs->FFT_SIZE) ;
for(k = 1 ; k <= pDnprefs->FFT_SIZE/2 ; k++) {
double freq = (double)pPrefs->rate / 2.0 /(double)(pDnprefs->FFT_SIZE/2)*(double)k ;
double db_left = 20.0*log10(left_noise_avg[k]/(max/2.0)) ;
double db_right = 20.0*log10(right_noise_avg[k]/(max/2.0)) ;
fprintf(fp, "%10lgHz %12.1lfdB %12.1lfdB\n", freq, db_left, db_right) ;
}
set_status_text(g_strconcat("noise sample written to ", g_get_current_dir(), "/noise.dat", NULL));
fclose(fp) ;
return 0 ;
}
void get_noise_sample(struct sound_prefs *pPrefs, struct denoise_prefs *pDnprefs, long noise_start, long noise_end,
fftw_real *left_noise_min, fftw_real *left_noise_max, fftw_real *left_noise_avg,
fftw_real *right_noise_min, fftw_real *right_noise_max, fftw_real *right_noise_avg)
{
int i, j, k ;
#ifdef HAVE_FFTW3
FFTW(plan) pForLeft, pForRight ;
#else /* HAVE_FFTW3 */
rfftw_plan pFor, pBak ;
#endif /* HAVE_FFTW3 */
fftw_real left[DENOISE_MAX_FFT], right[DENOISE_MAX_FFT] ;
fftw_real tmp[DENOISE_MAX_FFT] ;