forked from signalapp/Signal-Android
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpreprocess.c
1219 lines (1064 loc) · 40.7 KB
/
preprocess.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
/* Copyright (C) 2003 Epic Games (written by Jean-Marc Valin)
Copyright (C) 2004-2006 Epic Games
File: preprocess.c
Preprocessor with denoising based on the algorithm by Ephraim and Malah
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
1. Redistributions of source code must retain the above copyright notice,
this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
3. The name of the author may not be used to endorse or promote products
derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.
*/
/*
Recommended papers:
Y. Ephraim and D. Malah, "Speech enhancement using minimum mean-square error
short-time spectral amplitude estimator". IEEE Transactions on Acoustics,
Speech and Signal Processing, vol. ASSP-32, no. 6, pp. 1109-1121, 1984.
Y. Ephraim and D. Malah, "Speech enhancement using minimum mean-square error
log-spectral amplitude estimator". IEEE Transactions on Acoustics, Speech and
Signal Processing, vol. ASSP-33, no. 2, pp. 443-445, 1985.
I. Cohen and B. Berdugo, "Speech enhancement for non-stationary noise environments".
Signal Processing, vol. 81, no. 2, pp. 2403-2418, 2001.
Stefan Gustafsson, Rainer Martin, Peter Jax, and Peter Vary. "A psychoacoustic
approach to combined acoustic echo cancellation and noise reduction". IEEE
Transactions on Speech and Audio Processing, 2002.
J.-M. Valin, J. Rouat, and F. Michaud, "Microphone array post-filter for separation
of simultaneous non-stationary sources". In Proceedings IEEE International
Conference on Acoustics, Speech, and Signal Processing, 2004.
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include <math.h>
#include "speex/speex_preprocess.h"
#include "speex/speex_echo.h"
#include "arch.h"
#include "fftwrap.h"
#include "filterbank.h"
#include "math_approx.h"
#include "os_support.h"
#ifndef M_PI
#define M_PI 3.14159263
#endif
#define LOUDNESS_EXP 5.f
#define AMP_SCALE .001f
#define AMP_SCALE_1 1000.f
#define NB_BANDS 24
#define SPEECH_PROB_START_DEFAULT QCONST16(0.35f,15)
#define SPEECH_PROB_CONTINUE_DEFAULT QCONST16(0.20f,15)
#define NOISE_SUPPRESS_DEFAULT -15
#define ECHO_SUPPRESS_DEFAULT -40
#define ECHO_SUPPRESS_ACTIVE_DEFAULT -15
#ifndef NULL
#define NULL 0
#endif
#define SQR(x) ((x)*(x))
#define SQR16(x) (MULT16_16((x),(x)))
#define SQR16_Q15(x) (MULT16_16_Q15((x),(x)))
#ifdef FIXED_POINT
static inline spx_word16_t DIV32_16_Q8(spx_word32_t a, spx_word32_t b)
{
if (SHR32(a,7) >= b)
{
return 32767;
} else {
if (b>=QCONST32(1,23))
{
a = SHR32(a,8);
b = SHR32(b,8);
}
if (b>=QCONST32(1,19))
{
a = SHR32(a,4);
b = SHR32(b,4);
}
if (b>=QCONST32(1,15))
{
a = SHR32(a,4);
b = SHR32(b,4);
}
a = SHL32(a,8);
return PDIV32_16(a,b);
}
}
static inline spx_word16_t DIV32_16_Q15(spx_word32_t a, spx_word32_t b)
{
if (SHR32(a,15) >= b)
{
return 32767;
} else {
if (b>=QCONST32(1,23))
{
a = SHR32(a,8);
b = SHR32(b,8);
}
if (b>=QCONST32(1,19))
{
a = SHR32(a,4);
b = SHR32(b,4);
}
if (b>=QCONST32(1,15))
{
a = SHR32(a,4);
b = SHR32(b,4);
}
a = SHL32(a,15)-a;
return DIV32_16(a,b);
}
}
#define SNR_SCALING 256.f
#define SNR_SCALING_1 0.0039062f
#define SNR_SHIFT 8
#define FRAC_SCALING 32767.f
#define FRAC_SCALING_1 3.0518e-05
#define FRAC_SHIFT 1
#define EXPIN_SCALING 2048.f
#define EXPIN_SCALING_1 0.00048828f
#define EXPIN_SHIFT 11
#define EXPOUT_SCALING_1 1.5259e-05
#define NOISE_SHIFT 7
#else
#define DIV32_16_Q8(a,b) ((a)/(b))
#define DIV32_16_Q15(a,b) ((a)/(b))
#define SNR_SCALING 1.f
#define SNR_SCALING_1 1.f
#define SNR_SHIFT 0
#define FRAC_SCALING 1.f
#define FRAC_SCALING_1 1.f
#define FRAC_SHIFT 0
#define NOISE_SHIFT 0
#define EXPIN_SCALING 1.f
#define EXPIN_SCALING_1 1.f
#define EXPOUT_SCALING_1 1.f
#endif
/** Speex pre-processor state. */
struct SpeexPreprocessState_ {
/* Basic info */
int frame_size; /**< Number of samples processed each time */
int ps_size; /**< Number of points in the power spectrum */
int sampling_rate; /**< Sampling rate of the input/output */
int nbands;
FilterBank *bank;
/* Parameters */
int denoise_enabled;
int vad_enabled;
int dereverb_enabled;
spx_word16_t reverb_decay;
spx_word16_t reverb_level;
spx_word16_t speech_prob_start;
spx_word16_t speech_prob_continue;
int noise_suppress;
int echo_suppress;
int echo_suppress_active;
SpeexEchoState *echo_state;
spx_word16_t speech_prob; /**< Probability last frame was speech */
/* DSP-related arrays */
spx_word16_t *frame; /**< Processing frame (2*ps_size) */
spx_word16_t *ft; /**< Processing frame in freq domain (2*ps_size) */
spx_word32_t *ps; /**< Current power spectrum */
spx_word16_t *gain2; /**< Adjusted gains */
spx_word16_t *gain_floor; /**< Minimum gain allowed */
spx_word16_t *window; /**< Analysis/Synthesis window */
spx_word32_t *noise; /**< Noise estimate */
spx_word32_t *reverb_estimate; /**< Estimate of reverb energy */
spx_word32_t *old_ps; /**< Power spectrum for last frame */
spx_word16_t *gain; /**< Ephraim Malah gain */
spx_word16_t *prior; /**< A-priori SNR */
spx_word16_t *post; /**< A-posteriori SNR */
spx_word32_t *S; /**< Smoothed power spectrum */
spx_word32_t *Smin; /**< See Cohen paper */
spx_word32_t *Stmp; /**< See Cohen paper */
int *update_prob; /**< Probability of speech presence for noise update */
spx_word16_t *zeta; /**< Smoothed a priori SNR */
spx_word32_t *echo_noise;
spx_word32_t *residual_echo;
/* Misc */
spx_word16_t *inbuf; /**< Input buffer (overlapped analysis) */
spx_word16_t *outbuf; /**< Output buffer (for overlap and add) */
/* AGC stuff, only for floating point for now */
#ifndef FIXED_POINT
int agc_enabled;
float agc_level;
float loudness_accum;
float *loudness_weight; /**< Perceptual loudness curve */
float loudness; /**< Loudness estimate */
float agc_gain; /**< Current AGC gain */
float max_gain; /**< Maximum gain allowed */
float max_increase_step; /**< Maximum increase in gain from one frame to another */
float max_decrease_step; /**< Maximum decrease in gain from one frame to another */
float prev_loudness; /**< Loudness of previous frame */
float init_max; /**< Current gain limit during initialisation */
#endif
int nb_adapt; /**< Number of frames used for adaptation so far */
int was_speech;
int min_count; /**< Number of frames processed so far */
void *fft_lookup; /**< Lookup table for the FFT */
#ifdef FIXED_POINT
int frame_shift;
#endif
};
static void conj_window(spx_word16_t *w, int len)
{
int i;
for (i=0;i<len;i++)
{
spx_word16_t tmp;
#ifdef FIXED_POINT
spx_word16_t x = DIV32_16(MULT16_16(32767,i),len);
#else
spx_word16_t x = DIV32_16(MULT16_16(QCONST16(4.f,13),i),len);
#endif
int inv=0;
if (x<QCONST16(1.f,13))
{
} else if (x<QCONST16(2.f,13))
{
x=QCONST16(2.f,13)-x;
inv=1;
} else if (x<QCONST16(3.f,13))
{
x=x-QCONST16(2.f,13);
inv=1;
} else {
x=QCONST16(2.f,13)-x+QCONST16(2.f,13); /* 4 - x */
}
x = MULT16_16_Q14(QCONST16(1.271903f,14), x);
tmp = SQR16_Q15(QCONST16(.5f,15)-MULT16_16_P15(QCONST16(.5f,15),spx_cos_norm(SHL32(EXTEND32(x),2))));
if (inv)
tmp=SUB16(Q15_ONE,tmp);
w[i]=spx_sqrt(SHL32(EXTEND32(tmp),15));
}
}
#ifdef FIXED_POINT
/* This function approximates the gain function
y = gamma(1.25)^2 * M(-.25;1;-x) / sqrt(x)
which multiplied by xi/(1+xi) is the optimal gain
in the loudness domain ( sqrt[amplitude] )
Input in Q11 format, output in Q15
*/
static inline spx_word32_t hypergeom_gain(spx_word32_t xx)
{
int ind;
spx_word16_t frac;
/* Q13 table */
static const spx_word16_t table[21] = {
6730, 8357, 9868, 11267, 12563, 13770, 14898,
15959, 16961, 17911, 18816, 19682, 20512, 21311,
22082, 22827, 23549, 24250, 24931, 25594, 26241};
ind = SHR32(xx,10);
if (ind<0)
return Q15_ONE;
if (ind>19)
return ADD32(EXTEND32(Q15_ONE),EXTEND32(DIV32_16(QCONST32(.1296,23), SHR32(xx,EXPIN_SHIFT-SNR_SHIFT))));
frac = SHL32(xx-SHL32(ind,10),5);
return SHL32(DIV32_16(PSHR32(MULT16_16(Q15_ONE-frac,table[ind]) + MULT16_16(frac,table[ind+1]),7),(spx_sqrt(SHL32(xx,15)+6711))),7);
}
static inline spx_word16_t qcurve(spx_word16_t x)
{
x = MAX16(x, 1);
return DIV32_16(SHL32(EXTEND32(32767),9),ADD16(512,MULT16_16_Q15(QCONST16(.60f,15),DIV32_16(32767,x))));
}
/* Compute the gain floor based on different floors for the background noise and residual echo */
static void compute_gain_floor(int noise_suppress, int effective_echo_suppress, spx_word32_t *noise, spx_word32_t *echo, spx_word16_t *gain_floor, int len)
{
int i;
if (noise_suppress > effective_echo_suppress)
{
spx_word16_t noise_gain, gain_ratio;
noise_gain = EXTRACT16(MIN32(Q15_ONE,SHR32(spx_exp(MULT16_16(QCONST16(0.11513,11),noise_suppress)),1)));
gain_ratio = EXTRACT16(MIN32(Q15_ONE,SHR32(spx_exp(MULT16_16(QCONST16(.2302585f,11),effective_echo_suppress-noise_suppress)),1)));
/* gain_floor = sqrt [ (noise*noise_floor + echo*echo_floor) / (noise+echo) ] */
for (i=0;i<len;i++)
gain_floor[i] = MULT16_16_Q15(noise_gain,
spx_sqrt(SHL32(EXTEND32(DIV32_16_Q15(PSHR32(noise[i],NOISE_SHIFT) + MULT16_32_Q15(gain_ratio,echo[i]),
(1+PSHR32(noise[i],NOISE_SHIFT) + echo[i]) )),15)));
} else {
spx_word16_t echo_gain, gain_ratio;
echo_gain = EXTRACT16(MIN32(Q15_ONE,SHR32(spx_exp(MULT16_16(QCONST16(0.11513,11),effective_echo_suppress)),1)));
gain_ratio = EXTRACT16(MIN32(Q15_ONE,SHR32(spx_exp(MULT16_16(QCONST16(.2302585f,11),noise_suppress-effective_echo_suppress)),1)));
/* gain_floor = sqrt [ (noise*noise_floor + echo*echo_floor) / (noise+echo) ] */
for (i=0;i<len;i++)
gain_floor[i] = MULT16_16_Q15(echo_gain,
spx_sqrt(SHL32(EXTEND32(DIV32_16_Q15(MULT16_32_Q15(gain_ratio,PSHR32(noise[i],NOISE_SHIFT)) + echo[i],
(1+PSHR32(noise[i],NOISE_SHIFT) + echo[i]) )),15)));
}
}
#else
/* This function approximates the gain function
y = gamma(1.25)^2 * M(-.25;1;-x) / sqrt(x)
which multiplied by xi/(1+xi) is the optimal gain
in the loudness domain ( sqrt[amplitude] )
*/
static inline spx_word32_t hypergeom_gain(spx_word32_t xx)
{
int ind;
float integer, frac;
float x;
static const float table[21] = {
0.82157f, 1.02017f, 1.20461f, 1.37534f, 1.53363f, 1.68092f, 1.81865f,
1.94811f, 2.07038f, 2.18638f, 2.29688f, 2.40255f, 2.50391f, 2.60144f,
2.69551f, 2.78647f, 2.87458f, 2.96015f, 3.04333f, 3.12431f, 3.20326f};
x = EXPIN_SCALING_1*xx;
integer = floor(2*x);
ind = (int)integer;
if (ind<0)
return FRAC_SCALING;
if (ind>19)
return FRAC_SCALING*(1+.1296/x);
frac = 2*x-integer;
return FRAC_SCALING*((1-frac)*table[ind] + frac*table[ind+1])/sqrt(x+.0001f);
}
static inline spx_word16_t qcurve(spx_word16_t x)
{
return 1.f/(1.f+.15f/(SNR_SCALING_1*x));
}
static void compute_gain_floor(int noise_suppress, int effective_echo_suppress, spx_word32_t *noise, spx_word32_t *echo, spx_word16_t *gain_floor, int len)
{
int i;
float echo_floor;
float noise_floor;
noise_floor = exp(.2302585f*noise_suppress);
echo_floor = exp(.2302585f*effective_echo_suppress);
/* Compute the gain floor based on different floors for the background noise and residual echo */
for (i=0;i<len;i++)
gain_floor[i] = FRAC_SCALING*sqrt(noise_floor*PSHR32(noise[i],NOISE_SHIFT) + echo_floor*echo[i])/sqrt(1+PSHR32(noise[i],NOISE_SHIFT) + echo[i]);
}
#endif
EXPORT SpeexPreprocessState *speex_preprocess_state_init(int frame_size, int sampling_rate)
{
int i;
int N, N3, N4, M;
SpeexPreprocessState *st = (SpeexPreprocessState *)speex_alloc(sizeof(SpeexPreprocessState));
st->frame_size = frame_size;
/* Round ps_size down to the nearest power of two */
#if 0
i=1;
st->ps_size = st->frame_size;
while(1)
{
if (st->ps_size & ~i)
{
st->ps_size &= ~i;
i<<=1;
} else {
break;
}
}
if (st->ps_size < 3*st->frame_size/4)
st->ps_size = st->ps_size * 3 / 2;
#else
st->ps_size = st->frame_size;
#endif
N = st->ps_size;
N3 = 2*N - st->frame_size;
N4 = st->frame_size - N3;
st->sampling_rate = sampling_rate;
st->denoise_enabled = 1;
st->vad_enabled = 0;
st->dereverb_enabled = 0;
st->reverb_decay = 0;
st->reverb_level = 0;
st->noise_suppress = NOISE_SUPPRESS_DEFAULT;
st->echo_suppress = ECHO_SUPPRESS_DEFAULT;
st->echo_suppress_active = ECHO_SUPPRESS_ACTIVE_DEFAULT;
st->speech_prob_start = SPEECH_PROB_START_DEFAULT;
st->speech_prob_continue = SPEECH_PROB_CONTINUE_DEFAULT;
st->echo_state = NULL;
st->nbands = NB_BANDS;
M = st->nbands;
st->bank = filterbank_new(M, sampling_rate, N, 1);
st->frame = (spx_word16_t*)speex_alloc(2*N*sizeof(spx_word16_t));
st->window = (spx_word16_t*)speex_alloc(2*N*sizeof(spx_word16_t));
st->ft = (spx_word16_t*)speex_alloc(2*N*sizeof(spx_word16_t));
st->ps = (spx_word32_t*)speex_alloc((N+M)*sizeof(spx_word32_t));
st->noise = (spx_word32_t*)speex_alloc((N+M)*sizeof(spx_word32_t));
st->echo_noise = (spx_word32_t*)speex_alloc((N+M)*sizeof(spx_word32_t));
st->residual_echo = (spx_word32_t*)speex_alloc((N+M)*sizeof(spx_word32_t));
st->reverb_estimate = (spx_word32_t*)speex_alloc((N+M)*sizeof(spx_word32_t));
st->old_ps = (spx_word32_t*)speex_alloc((N+M)*sizeof(spx_word32_t));
st->prior = (spx_word16_t*)speex_alloc((N+M)*sizeof(spx_word16_t));
st->post = (spx_word16_t*)speex_alloc((N+M)*sizeof(spx_word16_t));
st->gain = (spx_word16_t*)speex_alloc((N+M)*sizeof(spx_word16_t));
st->gain2 = (spx_word16_t*)speex_alloc((N+M)*sizeof(spx_word16_t));
st->gain_floor = (spx_word16_t*)speex_alloc((N+M)*sizeof(spx_word16_t));
st->zeta = (spx_word16_t*)speex_alloc((N+M)*sizeof(spx_word16_t));
st->S = (spx_word32_t*)speex_alloc(N*sizeof(spx_word32_t));
st->Smin = (spx_word32_t*)speex_alloc(N*sizeof(spx_word32_t));
st->Stmp = (spx_word32_t*)speex_alloc(N*sizeof(spx_word32_t));
st->update_prob = (int*)speex_alloc(N*sizeof(int));
st->inbuf = (spx_word16_t*)speex_alloc(N3*sizeof(spx_word16_t));
st->outbuf = (spx_word16_t*)speex_alloc(N3*sizeof(spx_word16_t));
conj_window(st->window, 2*N3);
for (i=2*N3;i<2*st->ps_size;i++)
st->window[i]=Q15_ONE;
if (N4>0)
{
for (i=N3-1;i>=0;i--)
{
st->window[i+N3+N4]=st->window[i+N3];
st->window[i+N3]=1;
}
}
for (i=0;i<N+M;i++)
{
st->noise[i]=QCONST32(1.f,NOISE_SHIFT);
st->reverb_estimate[i]=0;
st->old_ps[i]=1;
st->gain[i]=Q15_ONE;
st->post[i]=SHL16(1, SNR_SHIFT);
st->prior[i]=SHL16(1, SNR_SHIFT);
}
for (i=0;i<N;i++)
st->update_prob[i] = 1;
for (i=0;i<N3;i++)
{
st->inbuf[i]=0;
st->outbuf[i]=0;
}
#ifndef FIXED_POINT
st->agc_enabled = 0;
st->agc_level = 8000;
st->loudness_weight = (float*)speex_alloc(N*sizeof(float));
for (i=0;i<N;i++)
{
float ff=((float)i)*.5*sampling_rate/((float)N);
/*st->loudness_weight[i] = .5f*(1.f/(1.f+ff/8000.f))+1.f*exp(-.5f*(ff-3800.f)*(ff-3800.f)/9e5f);*/
st->loudness_weight[i] = .35f-.35f*ff/16000.f+.73f*exp(-.5f*(ff-3800)*(ff-3800)/9e5f);
if (st->loudness_weight[i]<.01f)
st->loudness_weight[i]=.01f;
st->loudness_weight[i] *= st->loudness_weight[i];
}
/*st->loudness = pow(AMP_SCALE*st->agc_level,LOUDNESS_EXP);*/
st->loudness = 1e-15;
st->agc_gain = 1;
st->max_gain = 30;
st->max_increase_step = exp(0.11513f * 12.*st->frame_size / st->sampling_rate);
st->max_decrease_step = exp(-0.11513f * 40.*st->frame_size / st->sampling_rate);
st->prev_loudness = 1;
st->init_max = 1;
#endif
st->was_speech = 0;
st->fft_lookup = spx_fft_init(2*N);
st->nb_adapt=0;
st->min_count=0;
return st;
}
EXPORT void speex_preprocess_state_destroy(SpeexPreprocessState *st)
{
speex_free(st->frame);
speex_free(st->ft);
speex_free(st->ps);
speex_free(st->gain2);
speex_free(st->gain_floor);
speex_free(st->window);
speex_free(st->noise);
speex_free(st->reverb_estimate);
speex_free(st->old_ps);
speex_free(st->gain);
speex_free(st->prior);
speex_free(st->post);
#ifndef FIXED_POINT
speex_free(st->loudness_weight);
#endif
speex_free(st->echo_noise);
speex_free(st->residual_echo);
speex_free(st->S);
speex_free(st->Smin);
speex_free(st->Stmp);
speex_free(st->update_prob);
speex_free(st->zeta);
speex_free(st->inbuf);
speex_free(st->outbuf);
spx_fft_destroy(st->fft_lookup);
filterbank_destroy(st->bank);
speex_free(st);
}
/* FIXME: The AGC doesn't work yet with fixed-point*/
#ifndef FIXED_POINT
static void speex_compute_agc(SpeexPreprocessState *st, spx_word16_t Pframe, spx_word16_t *ft)
{
int i;
int N = st->ps_size;
float target_gain;
float loudness=1.f;
float rate;
for (i=2;i<N;i++)
{
loudness += 2.f*N*st->ps[i]* st->loudness_weight[i];
}
loudness=sqrt(loudness);
/*if (loudness < 2*pow(st->loudness, 1.0/LOUDNESS_EXP) &&
loudness*2 > pow(st->loudness, 1.0/LOUDNESS_EXP))*/
if (Pframe>.3f)
{
/*rate=2.0f*Pframe*Pframe/(1+st->nb_loudness_adapt);*/
rate = .03*Pframe*Pframe;
st->loudness = (1-rate)*st->loudness + (rate)*pow(AMP_SCALE*loudness, LOUDNESS_EXP);
st->loudness_accum = (1-rate)*st->loudness_accum + rate;
if (st->init_max < st->max_gain && st->nb_adapt > 20)
st->init_max *= 1.f + .1f*Pframe*Pframe;
}
/*printf ("%f %f %f %f\n", Pframe, loudness, pow(st->loudness, 1.0f/LOUDNESS_EXP), st->loudness2);*/
target_gain = AMP_SCALE*st->agc_level*pow(st->loudness/(1e-4+st->loudness_accum), -1.0f/LOUDNESS_EXP);
if ((Pframe>.5 && st->nb_adapt > 20) || target_gain < st->agc_gain)
{
if (target_gain > st->max_increase_step*st->agc_gain)
target_gain = st->max_increase_step*st->agc_gain;
if (target_gain < st->max_decrease_step*st->agc_gain && loudness < 10*st->prev_loudness)
target_gain = st->max_decrease_step*st->agc_gain;
if (target_gain > st->max_gain)
target_gain = st->max_gain;
if (target_gain > st->init_max)
target_gain = st->init_max;
st->agc_gain = target_gain;
}
/*fprintf (stderr, "%f %f %f\n", loudness, (float)AMP_SCALE_1*pow(st->loudness, 1.0f/LOUDNESS_EXP), st->agc_gain);*/
for (i=0;i<2*N;i++)
ft[i] *= st->agc_gain;
st->prev_loudness = loudness;
}
#endif
static void preprocess_analysis(SpeexPreprocessState *st, spx_int16_t *x)
{
int i;
int N = st->ps_size;
int N3 = 2*N - st->frame_size;
int N4 = st->frame_size - N3;
spx_word32_t *ps=st->ps;
/* 'Build' input frame */
for (i=0;i<N3;i++)
st->frame[i]=st->inbuf[i];
for (i=0;i<st->frame_size;i++)
st->frame[N3+i]=x[i];
/* Update inbuf */
for (i=0;i<N3;i++)
st->inbuf[i]=x[N4+i];
/* Windowing */
for (i=0;i<2*N;i++)
st->frame[i] = MULT16_16_Q15(st->frame[i], st->window[i]);
#ifdef FIXED_POINT
{
spx_word16_t max_val=0;
for (i=0;i<2*N;i++)
max_val = MAX16(max_val, ABS16(st->frame[i]));
st->frame_shift = 14-spx_ilog2(EXTEND32(max_val));
for (i=0;i<2*N;i++)
st->frame[i] = SHL16(st->frame[i], st->frame_shift);
}
#endif
/* Perform FFT */
spx_fft(st->fft_lookup, st->frame, st->ft);
/* Power spectrum */
ps[0]=MULT16_16(st->ft[0],st->ft[0]);
for (i=1;i<N;i++)
ps[i]=MULT16_16(st->ft[2*i-1],st->ft[2*i-1]) + MULT16_16(st->ft[2*i],st->ft[2*i]);
for (i=0;i<N;i++)
st->ps[i] = PSHR32(st->ps[i], 2*st->frame_shift);
filterbank_compute_bank32(st->bank, ps, ps+N);
}
static void update_noise_prob(SpeexPreprocessState *st)
{
int i;
int min_range;
int N = st->ps_size;
for (i=1;i<N-1;i++)
st->S[i] = MULT16_32_Q15(QCONST16(.8f,15),st->S[i]) + MULT16_32_Q15(QCONST16(.05f,15),st->ps[i-1])
+ MULT16_32_Q15(QCONST16(.1f,15),st->ps[i]) + MULT16_32_Q15(QCONST16(.05f,15),st->ps[i+1]);
st->S[0] = MULT16_32_Q15(QCONST16(.8f,15),st->S[0]) + MULT16_32_Q15(QCONST16(.2f,15),st->ps[0]);
st->S[N-1] = MULT16_32_Q15(QCONST16(.8f,15),st->S[N-1]) + MULT16_32_Q15(QCONST16(.2f,15),st->ps[N-1]);
if (st->nb_adapt==1)
{
for (i=0;i<N;i++)
st->Smin[i] = st->Stmp[i] = 0;
}
if (st->nb_adapt < 100)
min_range = 15;
else if (st->nb_adapt < 1000)
min_range = 50;
else if (st->nb_adapt < 10000)
min_range = 150;
else
min_range = 300;
if (st->min_count > min_range)
{
st->min_count = 0;
for (i=0;i<N;i++)
{
st->Smin[i] = MIN32(st->Stmp[i], st->S[i]);
st->Stmp[i] = st->S[i];
}
} else {
for (i=0;i<N;i++)
{
st->Smin[i] = MIN32(st->Smin[i], st->S[i]);
st->Stmp[i] = MIN32(st->Stmp[i], st->S[i]);
}
}
for (i=0;i<N;i++)
{
if (MULT16_32_Q15(QCONST16(.4f,15),st->S[i]) > st->Smin[i])
st->update_prob[i] = 1;
else
st->update_prob[i] = 0;
/*fprintf (stderr, "%f ", st->S[i]/st->Smin[i]);*/
/*fprintf (stderr, "%f ", st->update_prob[i]);*/
}
}
#define NOISE_OVERCOMPENS 1.
void speex_echo_get_residual(SpeexEchoState *st, spx_word32_t *Yout, int len);
EXPORT int speex_preprocess(SpeexPreprocessState *st, spx_int16_t *x, spx_int32_t *echo)
{
return speex_preprocess_run(st, x);
}
EXPORT int speex_preprocess_run(SpeexPreprocessState *st, spx_int16_t *x)
{
int i;
int M;
int N = st->ps_size;
int N3 = 2*N - st->frame_size;
int N4 = st->frame_size - N3;
spx_word32_t *ps=st->ps;
spx_word32_t Zframe;
spx_word16_t Pframe;
spx_word16_t beta, beta_1;
spx_word16_t effective_echo_suppress;
st->nb_adapt++;
if (st->nb_adapt>20000)
st->nb_adapt = 20000;
st->min_count++;
beta = MAX16(QCONST16(.03,15),DIV32_16(Q15_ONE,st->nb_adapt));
beta_1 = Q15_ONE-beta;
M = st->nbands;
/* Deal with residual echo if provided */
if (st->echo_state)
{
speex_echo_get_residual(st->echo_state, st->residual_echo, N);
#ifndef FIXED_POINT
/* If there are NaNs or ridiculous values, it'll show up in the DC and we just reset everything to zero */
if (!(st->residual_echo[0] >=0 && st->residual_echo[0]<N*1e9f))
{
for (i=0;i<N;i++)
st->residual_echo[i] = 0;
}
#endif
for (i=0;i<N;i++)
st->echo_noise[i] = MAX32(MULT16_32_Q15(QCONST16(.6f,15),st->echo_noise[i]), st->residual_echo[i]);
filterbank_compute_bank32(st->bank, st->echo_noise, st->echo_noise+N);
} else {
for (i=0;i<N+M;i++)
st->echo_noise[i] = 0;
}
preprocess_analysis(st, x);
update_noise_prob(st);
/* Noise estimation always updated for the 10 first frames */
/*if (st->nb_adapt<10)
{
for (i=1;i<N-1;i++)
st->update_prob[i] = 0;
}
*/
/* Update the noise estimate for the frequencies where it can be */
for (i=0;i<N;i++)
{
if (!st->update_prob[i] || st->ps[i] < PSHR32(st->noise[i], NOISE_SHIFT))
st->noise[i] = MAX32(EXTEND32(0),MULT16_32_Q15(beta_1,st->noise[i]) + MULT16_32_Q15(beta,SHL32(st->ps[i],NOISE_SHIFT)));
}
filterbank_compute_bank32(st->bank, st->noise, st->noise+N);
/* Special case for first frame */
if (st->nb_adapt==1)
for (i=0;i<N+M;i++)
st->old_ps[i] = ps[i];
/* Compute a posteriori SNR */
for (i=0;i<N+M;i++)
{
spx_word16_t gamma;
/* Total noise estimate including residual echo and reverberation */
spx_word32_t tot_noise = ADD32(ADD32(ADD32(EXTEND32(1), PSHR32(st->noise[i],NOISE_SHIFT)) , st->echo_noise[i]) , st->reverb_estimate[i]);
/* A posteriori SNR = ps/noise - 1*/
st->post[i] = SUB16(DIV32_16_Q8(ps[i],tot_noise), QCONST16(1.f,SNR_SHIFT));
st->post[i]=MIN16(st->post[i], QCONST16(100.f,SNR_SHIFT));
/* Computing update gamma = .1 + .9*(old/(old+noise))^2 */
gamma = QCONST16(.1f,15)+MULT16_16_Q15(QCONST16(.89f,15),SQR16_Q15(DIV32_16_Q15(st->old_ps[i],ADD32(st->old_ps[i],tot_noise))));
/* A priori SNR update = gamma*max(0,post) + (1-gamma)*old/noise */
st->prior[i] = EXTRACT16(PSHR32(ADD32(MULT16_16(gamma,MAX16(0,st->post[i])), MULT16_16(Q15_ONE-gamma,DIV32_16_Q8(st->old_ps[i],tot_noise))), 15));
st->prior[i]=MIN16(st->prior[i], QCONST16(100.f,SNR_SHIFT));
}
/*print_vec(st->post, N+M, "");*/
/* Recursive average of the a priori SNR. A bit smoothed for the psd components */
st->zeta[0] = PSHR32(ADD32(MULT16_16(QCONST16(.7f,15),st->zeta[0]), MULT16_16(QCONST16(.3f,15),st->prior[0])),15);
for (i=1;i<N-1;i++)
st->zeta[i] = PSHR32(ADD32(ADD32(ADD32(MULT16_16(QCONST16(.7f,15),st->zeta[i]), MULT16_16(QCONST16(.15f,15),st->prior[i])),
MULT16_16(QCONST16(.075f,15),st->prior[i-1])), MULT16_16(QCONST16(.075f,15),st->prior[i+1])),15);
for (i=N-1;i<N+M;i++)
st->zeta[i] = PSHR32(ADD32(MULT16_16(QCONST16(.7f,15),st->zeta[i]), MULT16_16(QCONST16(.3f,15),st->prior[i])),15);
/* Speech probability of presence for the entire frame is based on the average filterbank a priori SNR */
Zframe = 0;
for (i=N;i<N+M;i++)
Zframe = ADD32(Zframe, EXTEND32(st->zeta[i]));
Pframe = QCONST16(.1f,15)+MULT16_16_Q15(QCONST16(.899f,15),qcurve(DIV32_16(Zframe,st->nbands)));
effective_echo_suppress = EXTRACT16(PSHR32(ADD32(MULT16_16(SUB16(Q15_ONE,Pframe), st->echo_suppress), MULT16_16(Pframe, st->echo_suppress_active)),15));
compute_gain_floor(st->noise_suppress, effective_echo_suppress, st->noise+N, st->echo_noise+N, st->gain_floor+N, M);
/* Compute Ephraim & Malah gain speech probability of presence for each critical band (Bark scale)
Technically this is actually wrong because the EM gaim assumes a slightly different probability
distribution */
for (i=N;i<N+M;i++)
{
/* See EM and Cohen papers*/
spx_word32_t theta;
/* Gain from hypergeometric function */
spx_word32_t MM;
/* Weiner filter gain */
spx_word16_t prior_ratio;
/* a priority probability of speech presence based on Bark sub-band alone */
spx_word16_t P1;
/* Speech absence a priori probability (considering sub-band and frame) */
spx_word16_t q;
#ifdef FIXED_POINT
spx_word16_t tmp;
#endif
prior_ratio = PDIV32_16(SHL32(EXTEND32(st->prior[i]), 15), ADD16(st->prior[i], SHL32(1,SNR_SHIFT)));
theta = MULT16_32_P15(prior_ratio, QCONST32(1.f,EXPIN_SHIFT)+SHL32(EXTEND32(st->post[i]),EXPIN_SHIFT-SNR_SHIFT));
MM = hypergeom_gain(theta);
/* Gain with bound */
st->gain[i] = EXTRACT16(MIN32(Q15_ONE, MULT16_32_Q15(prior_ratio, MM)));
/* Save old Bark power spectrum */
st->old_ps[i] = MULT16_32_P15(QCONST16(.2f,15),st->old_ps[i]) + MULT16_32_P15(MULT16_16_P15(QCONST16(.8f,15),SQR16_Q15(st->gain[i])),ps[i]);
P1 = QCONST16(.199f,15)+MULT16_16_Q15(QCONST16(.8f,15),qcurve (st->zeta[i]));
q = Q15_ONE-MULT16_16_Q15(Pframe,P1);
#ifdef FIXED_POINT
theta = MIN32(theta, EXTEND32(32767));
/*Q8*/tmp = MULT16_16_Q15((SHL32(1,SNR_SHIFT)+st->prior[i]),EXTRACT16(MIN32(Q15ONE,SHR32(spx_exp(-EXTRACT16(theta)),1))));
tmp = MIN16(QCONST16(3.,SNR_SHIFT), tmp); /* Prevent overflows in the next line*/
/*Q8*/tmp = EXTRACT16(PSHR32(MULT16_16(PDIV32_16(SHL32(EXTEND32(q),8),(Q15_ONE-q)),tmp),8));
st->gain2[i]=DIV32_16(SHL32(EXTEND32(32767),SNR_SHIFT), ADD16(256,tmp));
#else
st->gain2[i]=1/(1.f + (q/(1.f-q))*(1+st->prior[i])*exp(-theta));
#endif
}
/* Convert the EM gains and speech prob to linear frequency */
filterbank_compute_psd16(st->bank,st->gain2+N, st->gain2);
filterbank_compute_psd16(st->bank,st->gain+N, st->gain);
/* Use 1 for linear gain resolution (best) or 0 for Bark gain resolution (faster) */
if (1)
{
filterbank_compute_psd16(st->bank,st->gain_floor+N, st->gain_floor);
/* Compute gain according to the Ephraim-Malah algorithm -- linear frequency */
for (i=0;i<N;i++)
{
spx_word32_t MM;
spx_word32_t theta;
spx_word16_t prior_ratio;
spx_word16_t tmp;
spx_word16_t p;
spx_word16_t g;
/* Wiener filter gain */
prior_ratio = PDIV32_16(SHL32(EXTEND32(st->prior[i]), 15), ADD16(st->prior[i], SHL32(1,SNR_SHIFT)));
theta = MULT16_32_P15(prior_ratio, QCONST32(1.f,EXPIN_SHIFT)+SHL32(EXTEND32(st->post[i]),EXPIN_SHIFT-SNR_SHIFT));
/* Optimal estimator for loudness domain */
MM = hypergeom_gain(theta);
/* EM gain with bound */
g = EXTRACT16(MIN32(Q15_ONE, MULT16_32_Q15(prior_ratio, MM)));
/* Interpolated speech probability of presence */
p = st->gain2[i];
/* Constrain the gain to be close to the Bark scale gain */
if (MULT16_16_Q15(QCONST16(.333f,15),g) > st->gain[i])
g = MULT16_16(3,st->gain[i]);
st->gain[i] = g;
/* Save old power spectrum */
st->old_ps[i] = MULT16_32_P15(QCONST16(.2f,15),st->old_ps[i]) + MULT16_32_P15(MULT16_16_P15(QCONST16(.8f,15),SQR16_Q15(st->gain[i])),ps[i]);
/* Apply gain floor */
if (st->gain[i] < st->gain_floor[i])
st->gain[i] = st->gain_floor[i];
/* Exponential decay model for reverberation (unused) */
/*st->reverb_estimate[i] = st->reverb_decay*st->reverb_estimate[i] + st->reverb_decay*st->reverb_level*st->gain[i]*st->gain[i]*st->ps[i];*/
/* Take into account speech probability of presence (loudness domain MMSE estimator) */
/* gain2 = [p*sqrt(gain)+(1-p)*sqrt(gain _floor) ]^2 */
tmp = MULT16_16_P15(p,spx_sqrt(SHL32(EXTEND32(st->gain[i]),15))) + MULT16_16_P15(SUB16(Q15_ONE,p),spx_sqrt(SHL32(EXTEND32(st->gain_floor[i]),15)));
st->gain2[i]=SQR16_Q15(tmp);
/* Use this if you want a log-domain MMSE estimator instead */
/*st->gain2[i] = pow(st->gain[i], p) * pow(st->gain_floor[i],1.f-p);*/
}
} else {
for (i=N;i<N+M;i++)
{
spx_word16_t tmp;
spx_word16_t p = st->gain2[i];
st->gain[i] = MAX16(st->gain[i], st->gain_floor[i]);
tmp = MULT16_16_P15(p,spx_sqrt(SHL32(EXTEND32(st->gain[i]),15))) + MULT16_16_P15(SUB16(Q15_ONE,p),spx_sqrt(SHL32(EXTEND32(st->gain_floor[i]),15)));
st->gain2[i]=SQR16_Q15(tmp);
}
filterbank_compute_psd16(st->bank,st->gain2+N, st->gain2);
}
/* If noise suppression is off, don't apply the gain (but then why call this in the first place!) */
if (!st->denoise_enabled)
{
for (i=0;i<N+M;i++)
st->gain2[i]=Q15_ONE;
}
/* Apply computed gain */
for (i=1;i<N;i++)
{
st->ft[2*i-1] = MULT16_16_P15(st->gain2[i],st->ft[2*i-1]);
st->ft[2*i] = MULT16_16_P15(st->gain2[i],st->ft[2*i]);
}
st->ft[0] = MULT16_16_P15(st->gain2[0],st->ft[0]);
st->ft[2*N-1] = MULT16_16_P15(st->gain2[N-1],st->ft[2*N-1]);
/*FIXME: This *will* not work for fixed-point */
#ifndef FIXED_POINT
if (st->agc_enabled)
speex_compute_agc(st, Pframe, st->ft);
#endif
/* Inverse FFT with 1/N scaling */
spx_ifft(st->fft_lookup, st->ft, st->frame);
/* Scale back to original (lower) amplitude */
for (i=0;i<2*N;i++)
st->frame[i] = PSHR16(st->frame[i], st->frame_shift);
/*FIXME: This *will* not work for fixed-point */
#ifndef FIXED_POINT
if (st->agc_enabled)
{
float max_sample=0;
for (i=0;i<2*N;i++)
if (fabs(st->frame[i])>max_sample)
max_sample = fabs(st->frame[i]);
if (max_sample>28000.f)
{
float damp = 28000.f/max_sample;
for (i=0;i<2*N;i++)
st->frame[i] *= damp;
}
}
#endif
/* Synthesis window (for WOLA) */
for (i=0;i<2*N;i++)
st->frame[i] = MULT16_16_Q15(st->frame[i], st->window[i]);
/* Perform overlap and add */
for (i=0;i<N3;i++)
x[i] = st->outbuf[i] + st->frame[i];
for (i=0;i<N4;i++)
x[N3+i] = st->frame[N3+i];
/* Update outbuf */
for (i=0;i<N3;i++)
st->outbuf[i] = st->frame[st->frame_size+i];
/* FIXME: This VAD is a kludge */
st->speech_prob = Pframe;
if (st->vad_enabled)
{
if (st->speech_prob > st->speech_prob_start || (st->was_speech && st->speech_prob > st->speech_prob_continue))
{