1 /*
2  *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
3  *
4  *  Use of this source code is governed by a BSD-style license
5  *  that can be found in the LICENSE file in the root of the source
6  *  tree. An additional intellectual property rights grant can be found
7  *  in the file PATENTS.  All contributing project authors may
8  *  be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 #include "modules/audio_processing/agc/legacy/digital_agc.h"
12 
13 #include <string.h>
14 
15 #include "modules/audio_processing/agc/legacy/gain_control.h"
16 #include "rtc_base/checks.h"
17 
18 namespace webrtc {
19 
20 namespace {
21 
22 // To generate the gaintable, copy&paste the following lines to a Matlab window:
23 // MaxGain = 6; MinGain = 0; CompRatio = 3; Knee = 1;
24 // zeros = 0:31; lvl = 2.^(1-zeros);
25 // A = -10*log10(lvl) * (CompRatio - 1) / CompRatio;
26 // B = MaxGain - MinGain;
27 // gains = round(2^16*10.^(0.05 * (MinGain + B * (
28 // log(exp(-Knee*A)+exp(-Knee*B)) - log(1+exp(-Knee*B)) ) /
29 // log(1/(1+exp(Knee*B))))));
30 // fprintf(1, '\t%i, %i, %i, %i,\n', gains);
31 // % Matlab code for plotting the gain and input/output level characteristic
32 // (copy/paste the following 3 lines):
33 // in = 10*log10(lvl); out = 20*log10(gains/65536);
34 // subplot(121); plot(in, out); axis([-30, 0, -5, 20]); grid on; xlabel('Input
35 // (dB)'); ylabel('Gain (dB)');
36 // subplot(122); plot(in, in+out); axis([-30, 0, -30, 5]); grid on;
37 // xlabel('Input (dB)'); ylabel('Output (dB)');
38 // zoom on;
39 
40 // Generator table for y=log2(1+e^x) in Q8.
41 enum { kGenFuncTableSize = 128 };
42 static const uint16_t kGenFuncTable[kGenFuncTableSize] = {
43     256,   485,   786,   1126,  1484,  1849,  2217,  2586,  2955,  3324,  3693,
44     4063,  4432,  4801,  5171,  5540,  5909,  6279,  6648,  7017,  7387,  7756,
45     8125,  8495,  8864,  9233,  9603,  9972,  10341, 10711, 11080, 11449, 11819,
46     12188, 12557, 12927, 13296, 13665, 14035, 14404, 14773, 15143, 15512, 15881,
47     16251, 16620, 16989, 17359, 17728, 18097, 18466, 18836, 19205, 19574, 19944,
48     20313, 20682, 21052, 21421, 21790, 22160, 22529, 22898, 23268, 23637, 24006,
49     24376, 24745, 25114, 25484, 25853, 26222, 26592, 26961, 27330, 27700, 28069,
50     28438, 28808, 29177, 29546, 29916, 30285, 30654, 31024, 31393, 31762, 32132,
51     32501, 32870, 33240, 33609, 33978, 34348, 34717, 35086, 35456, 35825, 36194,
52     36564, 36933, 37302, 37672, 38041, 38410, 38780, 39149, 39518, 39888, 40257,
53     40626, 40996, 41365, 41734, 42104, 42473, 42842, 43212, 43581, 43950, 44320,
54     44689, 45058, 45428, 45797, 46166, 46536, 46905};
55 
56 static const int16_t kAvgDecayTime = 250;  // frames; < 3000
57 
58 // the 32 most significant bits of A(19) * B(26) >> 13
59 #define AGC_MUL32(A, B) (((B) >> 13) * (A) + (((0x00001FFF & (B)) * (A)) >> 13))
60 // C + the 32 most significant bits of A * B
61 #define AGC_SCALEDIFF32(A, B, C) \
62   ((C) + ((B) >> 16) * (A) + (((0x0000FFFF & (B)) * (A)) >> 16))
63 
64 }  // namespace
65 
WebRtcAgc_CalculateGainTable(int32_t * gainTable,int16_t digCompGaindB,int16_t targetLevelDbfs,uint8_t limiterEnable,int16_t analogTarget)66 int32_t WebRtcAgc_CalculateGainTable(int32_t* gainTable,       // Q16
67                                      int16_t digCompGaindB,    // Q0
68                                      int16_t targetLevelDbfs,  // Q0
69                                      uint8_t limiterEnable,
70                                      int16_t analogTarget) {  // Q0
71   // This function generates the compressor gain table used in the fixed digital
72   // part.
73   uint32_t tmpU32no1, tmpU32no2, absInLevel, logApprox;
74   int32_t inLevel, limiterLvl;
75   int32_t tmp32, tmp32no1, tmp32no2, numFIX, den, y32;
76   const uint16_t kLog10 = 54426;    // log2(10)     in Q14
77   const uint16_t kLog10_2 = 49321;  // 10*log10(2)  in Q14
78   const uint16_t kLogE_1 = 23637;   // log2(e)      in Q14
79   uint16_t constMaxGain;
80   uint16_t tmpU16, intPart, fracPart;
81   const int16_t kCompRatio = 3;
82   const int16_t kSoftLimiterLeft = 1;
83   int16_t limiterOffset = 0;  // Limiter offset
84   int16_t limiterIdx, limiterLvlX;
85   int16_t constLinApprox, zeroGainLvl, maxGain, diffGain;
86   int16_t i, tmp16, tmp16no1;
87   int zeros, zerosScale;
88 
89   // Constants
90   //    kLogE_1 = 23637; // log2(e)      in Q14
91   //    kLog10 = 54426; // log2(10)     in Q14
92   //    kLog10_2 = 49321; // 10*log10(2)  in Q14
93 
94   // Calculate maximum digital gain and zero gain level
95   tmp32no1 = (digCompGaindB - analogTarget) * (kCompRatio - 1);
96   tmp16no1 = analogTarget - targetLevelDbfs;
97   tmp16no1 +=
98       WebRtcSpl_DivW32W16ResW16(tmp32no1 + (kCompRatio >> 1), kCompRatio);
99   maxGain = WEBRTC_SPL_MAX(tmp16no1, (analogTarget - targetLevelDbfs));
100   tmp32no1 = maxGain * kCompRatio;
101   zeroGainLvl = digCompGaindB;
102   zeroGainLvl -= WebRtcSpl_DivW32W16ResW16(tmp32no1 + ((kCompRatio - 1) >> 1),
103                                            kCompRatio - 1);
104   if ((digCompGaindB <= analogTarget) && (limiterEnable)) {
105     zeroGainLvl += (analogTarget - digCompGaindB + kSoftLimiterLeft);
106     limiterOffset = 0;
107   }
108 
109   // Calculate the difference between maximum gain and gain at 0dB0v:
110   //  diffGain = maxGain + (compRatio-1)*zeroGainLvl/compRatio
111   //           = (compRatio-1)*digCompGaindB/compRatio
112   tmp32no1 = digCompGaindB * (kCompRatio - 1);
113   diffGain =
114       WebRtcSpl_DivW32W16ResW16(tmp32no1 + (kCompRatio >> 1), kCompRatio);
115   if (diffGain < 0 || diffGain >= kGenFuncTableSize) {
116     RTC_DCHECK(0);
117     return -1;
118   }
119 
120   // Calculate the limiter level and index:
121   //  limiterLvlX = analogTarget - limiterOffset
122   //  limiterLvl  = targetLevelDbfs + limiterOffset/compRatio
123   limiterLvlX = analogTarget - limiterOffset;
124   limiterIdx = 2 + WebRtcSpl_DivW32W16ResW16((int32_t)limiterLvlX * (1 << 13),
125                                              kLog10_2 / 2);
126   tmp16no1 =
127       WebRtcSpl_DivW32W16ResW16(limiterOffset + (kCompRatio >> 1), kCompRatio);
128   limiterLvl = targetLevelDbfs + tmp16no1;
129 
130   // Calculate (through table lookup):
131   //  constMaxGain = log2(1+2^(log2(e)*diffGain)); (in Q8)
132   constMaxGain = kGenFuncTable[diffGain];  // in Q8
133 
134   // Calculate a parameter used to approximate the fractional part of 2^x with a
135   // piecewise linear function in Q14:
136   //  constLinApprox = round(3/2*(4*(3-2*sqrt(2))/(log(2)^2)-0.5)*2^14);
137   constLinApprox = 22817;  // in Q14
138 
139   // Calculate a denominator used in the exponential part to convert from dB to
140   // linear scale:
141   //  den = 20*constMaxGain (in Q8)
142   den = WEBRTC_SPL_MUL_16_U16(20, constMaxGain);  // in Q8
143 
144   for (i = 0; i < 32; i++) {
145     // Calculate scaled input level (compressor):
146     //  inLevel =
147     //  fix((-constLog10_2*(compRatio-1)*(1-i)+fix(compRatio/2))/compRatio)
148     tmp16 = (int16_t)((kCompRatio - 1) * (i - 1));       // Q0
149     tmp32 = WEBRTC_SPL_MUL_16_U16(tmp16, kLog10_2) + 1;  // Q14
150     inLevel = WebRtcSpl_DivW32W16(tmp32, kCompRatio);    // Q14
151 
152     // Calculate diffGain-inLevel, to map using the genFuncTable
153     inLevel = (int32_t)diffGain * (1 << 14) - inLevel;  // Q14
154 
155     // Make calculations on abs(inLevel) and compensate for the sign afterwards.
156     absInLevel = (uint32_t)WEBRTC_SPL_ABS_W32(inLevel);  // Q14
157 
158     // LUT with interpolation
159     intPart = (uint16_t)(absInLevel >> 14);
160     fracPart =
161         (uint16_t)(absInLevel & 0x00003FFF);  // extract the fractional part
162     tmpU16 = kGenFuncTable[intPart + 1] - kGenFuncTable[intPart];  // Q8
163     tmpU32no1 = tmpU16 * fracPart;                                 // Q22
164     tmpU32no1 += (uint32_t)kGenFuncTable[intPart] << 14;           // Q22
165     logApprox = tmpU32no1 >> 8;                                    // Q14
166     // Compensate for negative exponent using the relation:
167     //  log2(1 + 2^-x) = log2(1 + 2^x) - x
168     if (inLevel < 0) {
169       zeros = WebRtcSpl_NormU32(absInLevel);
170       zerosScale = 0;
171       if (zeros < 15) {
172         // Not enough space for multiplication
173         tmpU32no2 = absInLevel >> (15 - zeros);                 // Q(zeros-1)
174         tmpU32no2 = WEBRTC_SPL_UMUL_32_16(tmpU32no2, kLogE_1);  // Q(zeros+13)
175         if (zeros < 9) {
176           zerosScale = 9 - zeros;
177           tmpU32no1 >>= zerosScale;  // Q(zeros+13)
178         } else {
179           tmpU32no2 >>= zeros - 9;  // Q22
180         }
181       } else {
182         tmpU32no2 = WEBRTC_SPL_UMUL_32_16(absInLevel, kLogE_1);  // Q28
183         tmpU32no2 >>= 6;                                         // Q22
184       }
185       logApprox = 0;
186       if (tmpU32no2 < tmpU32no1) {
187         logApprox = (tmpU32no1 - tmpU32no2) >> (8 - zerosScale);  // Q14
188       }
189     }
190     numFIX = (maxGain * constMaxGain) * (1 << 6);  // Q14
191     numFIX -= (int32_t)logApprox * diffGain;       // Q14
192 
193     // Calculate ratio
194     // Shift |numFIX| as much as possible.
195     // Ensure we avoid wrap-around in |den| as well.
196     if (numFIX > (den >> 8) || -numFIX > (den >> 8)) {  // |den| is Q8.
197       zeros = WebRtcSpl_NormW32(numFIX);
198     } else {
199       zeros = WebRtcSpl_NormW32(den) + 8;
200     }
201     numFIX *= 1 << zeros;  // Q(14+zeros)
202 
203     // Shift den so we end up in Qy1
204     tmp32no1 = WEBRTC_SPL_SHIFT_W32(den, zeros - 9);  // Q(zeros - 1)
205     y32 = numFIX / tmp32no1;                          // in Q15
206     // This is to do rounding in Q14.
207     y32 = y32 >= 0 ? (y32 + 1) >> 1 : -((-y32 + 1) >> 1);
208 
209     if (limiterEnable && (i < limiterIdx)) {
210       tmp32 = WEBRTC_SPL_MUL_16_U16(i - 1, kLog10_2);  // Q14
211       tmp32 -= limiterLvl * (1 << 14);                 // Q14
212       y32 = WebRtcSpl_DivW32W16(tmp32 + 10, 20);
213     }
214     if (y32 > 39000) {
215       tmp32 = (y32 >> 1) * kLog10 + 4096;  // in Q27
216       tmp32 >>= 13;                        // In Q14.
217     } else {
218       tmp32 = y32 * kLog10 + 8192;  // in Q28
219       tmp32 >>= 14;                 // In Q14.
220     }
221     tmp32 += 16 << 14;  // in Q14 (Make sure final output is in Q16)
222 
223     // Calculate power
224     if (tmp32 > 0) {
225       intPart = (int16_t)(tmp32 >> 14);
226       fracPart = (uint16_t)(tmp32 & 0x00003FFF);  // in Q14
227       if ((fracPart >> 13) != 0) {
228         tmp16 = (2 << 14) - constLinApprox;
229         tmp32no2 = (1 << 14) - fracPart;
230         tmp32no2 *= tmp16;
231         tmp32no2 >>= 13;
232         tmp32no2 = (1 << 14) - tmp32no2;
233       } else {
234         tmp16 = constLinApprox - (1 << 14);
235         tmp32no2 = (fracPart * tmp16) >> 13;
236       }
237       fracPart = (uint16_t)tmp32no2;
238       gainTable[i] =
239           (1 << intPart) + WEBRTC_SPL_SHIFT_W32(fracPart, intPart - 14);
240     } else {
241       gainTable[i] = 0;
242     }
243   }
244 
245   return 0;
246 }
247 
WebRtcAgc_InitDigital(DigitalAgc * stt,int16_t agcMode)248 int32_t WebRtcAgc_InitDigital(DigitalAgc* stt, int16_t agcMode) {
249   if (agcMode == kAgcModeFixedDigital) {
250     // start at minimum to find correct gain faster
251     stt->capacitorSlow = 0;
252   } else {
253     // start out with 0 dB gain
254     stt->capacitorSlow = 134217728;  // (int32_t)(0.125f * 32768.0f * 32768.0f);
255   }
256   stt->capacitorFast = 0;
257   stt->gain = 65536;
258   stt->gatePrevious = 0;
259   stt->agcMode = agcMode;
260 
261   // initialize VADs
262   WebRtcAgc_InitVad(&stt->vadNearend);
263   WebRtcAgc_InitVad(&stt->vadFarend);
264 
265   return 0;
266 }
267 
WebRtcAgc_AddFarendToDigital(DigitalAgc * stt,const int16_t * in_far,size_t nrSamples)268 int32_t WebRtcAgc_AddFarendToDigital(DigitalAgc* stt,
269                                      const int16_t* in_far,
270                                      size_t nrSamples) {
271   RTC_DCHECK(stt);
272   // VAD for far end
273   WebRtcAgc_ProcessVad(&stt->vadFarend, in_far, nrSamples);
274 
275   return 0;
276 }
277 
278 // Gains is an 11 element long array (one value per ms, incl start & end).
WebRtcAgc_ComputeDigitalGains(DigitalAgc * stt,const int16_t * const * in_near,size_t num_bands,uint32_t FS,int16_t lowlevelSignal,int32_t gains[11])279 int32_t WebRtcAgc_ComputeDigitalGains(DigitalAgc* stt,
280                                       const int16_t* const* in_near,
281                                       size_t num_bands,
282                                       uint32_t FS,
283                                       int16_t lowlevelSignal,
284                                       int32_t gains[11]) {
285   int32_t tmp32;
286   int32_t env[10];
287   int32_t max_nrg;
288   int32_t cur_level;
289   int32_t gain32;
290   int16_t logratio;
291   int16_t lower_thr, upper_thr;
292   int16_t zeros = 0, zeros_fast, frac = 0;
293   int16_t decay;
294   int16_t gate, gain_adj;
295   int16_t k;
296   size_t n, L;
297   int16_t L2;  // samples/subframe
298 
299   // determine number of samples per ms
300   if (FS == 8000) {
301     L = 8;
302     L2 = 3;
303   } else if (FS == 16000 || FS == 32000 || FS == 48000) {
304     L = 16;
305     L2 = 4;
306   } else {
307     return -1;
308   }
309 
310   // VAD for near end
311   logratio = WebRtcAgc_ProcessVad(&stt->vadNearend, in_near[0], L * 10);
312 
313   // Account for far end VAD
314   if (stt->vadFarend.counter > 10) {
315     tmp32 = 3 * logratio;
316     logratio = (int16_t)((tmp32 - stt->vadFarend.logRatio) >> 2);
317   }
318 
319   // Determine decay factor depending on VAD
320   //  upper_thr = 1.0f;
321   //  lower_thr = 0.25f;
322   upper_thr = 1024;  // Q10
323   lower_thr = 0;     // Q10
324   if (logratio > upper_thr) {
325     // decay = -2^17 / DecayTime;  ->  -65
326     decay = -65;
327   } else if (logratio < lower_thr) {
328     decay = 0;
329   } else {
330     // decay = (int16_t)(((lower_thr - logratio)
331     //       * (2^27/(DecayTime*(upper_thr-lower_thr)))) >> 10);
332     // SUBSTITUTED: 2^27/(DecayTime*(upper_thr-lower_thr))  ->  65
333     tmp32 = (lower_thr - logratio) * 65;
334     decay = (int16_t)(tmp32 >> 10);
335   }
336 
337   // adjust decay factor for long silence (detected as low standard deviation)
338   // This is only done in the adaptive modes
339   if (stt->agcMode != kAgcModeFixedDigital) {
340     if (stt->vadNearend.stdLongTerm < 4000) {
341       decay = 0;
342     } else if (stt->vadNearend.stdLongTerm < 8096) {
343       // decay = (int16_t)(((stt->vadNearend.stdLongTerm - 4000) * decay) >>
344       // 12);
345       tmp32 = (stt->vadNearend.stdLongTerm - 4000) * decay;
346       decay = (int16_t)(tmp32 >> 12);
347     }
348 
349     if (lowlevelSignal != 0) {
350       decay = 0;
351     }
352   }
353   // Find max amplitude per sub frame
354   // iterate over sub frames
355   for (k = 0; k < 10; k++) {
356     // iterate over samples
357     max_nrg = 0;
358     for (n = 0; n < L; n++) {
359       int32_t nrg = in_near[0][k * L + n] * in_near[0][k * L + n];
360       if (nrg > max_nrg) {
361         max_nrg = nrg;
362       }
363     }
364     env[k] = max_nrg;
365   }
366 
367   // Calculate gain per sub frame
368   gains[0] = stt->gain;
369   for (k = 0; k < 10; k++) {
370     // Fast envelope follower
371     //  decay time = -131000 / -1000 = 131 (ms)
372     stt->capacitorFast =
373         AGC_SCALEDIFF32(-1000, stt->capacitorFast, stt->capacitorFast);
374     if (env[k] > stt->capacitorFast) {
375       stt->capacitorFast = env[k];
376     }
377     // Slow envelope follower
378     if (env[k] > stt->capacitorSlow) {
379       // increase capacitorSlow
380       stt->capacitorSlow = AGC_SCALEDIFF32(500, (env[k] - stt->capacitorSlow),
381                                            stt->capacitorSlow);
382     } else {
383       // decrease capacitorSlow
384       stt->capacitorSlow =
385           AGC_SCALEDIFF32(decay, stt->capacitorSlow, stt->capacitorSlow);
386     }
387 
388     // use maximum of both capacitors as current level
389     if (stt->capacitorFast > stt->capacitorSlow) {
390       cur_level = stt->capacitorFast;
391     } else {
392       cur_level = stt->capacitorSlow;
393     }
394     // Translate signal level into gain, using a piecewise linear approximation
395     // find number of leading zeros
396     zeros = WebRtcSpl_NormU32((uint32_t)cur_level);
397     if (cur_level == 0) {
398       zeros = 31;
399     }
400     tmp32 = ((uint32_t)cur_level << zeros) & 0x7FFFFFFF;
401     frac = (int16_t)(tmp32 >> 19);  // Q12.
402     // Interpolate between gainTable[zeros] and gainTable[zeros-1].
403     tmp32 =
404         ((stt->gainTable[zeros - 1] - stt->gainTable[zeros]) * (int64_t)frac) >>
405         12;
406     gains[k + 1] = stt->gainTable[zeros] + tmp32;
407   }
408 
409   // Gate processing (lower gain during absence of speech)
410   zeros = (zeros << 9) - (frac >> 3);
411   // find number of leading zeros
412   zeros_fast = WebRtcSpl_NormU32((uint32_t)stt->capacitorFast);
413   if (stt->capacitorFast == 0) {
414     zeros_fast = 31;
415   }
416   tmp32 = ((uint32_t)stt->capacitorFast << zeros_fast) & 0x7FFFFFFF;
417   zeros_fast <<= 9;
418   zeros_fast -= (int16_t)(tmp32 >> 22);
419 
420   gate = 1000 + zeros_fast - zeros - stt->vadNearend.stdShortTerm;
421 
422   if (gate < 0) {
423     stt->gatePrevious = 0;
424   } else {
425     tmp32 = stt->gatePrevious * 7;
426     gate = (int16_t)((gate + tmp32) >> 3);
427     stt->gatePrevious = gate;
428   }
429   // gate < 0     -> no gate
430   // gate > 2500  -> max gate
431   if (gate > 0) {
432     if (gate < 2500) {
433       gain_adj = (2500 - gate) >> 5;
434     } else {
435       gain_adj = 0;
436     }
437     for (k = 0; k < 10; k++) {
438       if ((gains[k + 1] - stt->gainTable[0]) > 8388608) {
439         // To prevent wraparound
440         tmp32 = (gains[k + 1] - stt->gainTable[0]) >> 8;
441         tmp32 *= 178 + gain_adj;
442       } else {
443         tmp32 = (gains[k + 1] - stt->gainTable[0]) * (178 + gain_adj);
444         tmp32 >>= 8;
445       }
446       gains[k + 1] = stt->gainTable[0] + tmp32;
447     }
448   }
449 
450   // Limit gain to avoid overload distortion
451   for (k = 0; k < 10; k++) {
452     // Find a shift of gains[k + 1] such that it can be squared without
453     // overflow, but at least by 10 bits.
454     zeros = 10;
455     if (gains[k + 1] > 47452159) {
456       zeros = 16 - WebRtcSpl_NormW32(gains[k + 1]);
457     }
458     gain32 = (gains[k + 1] >> zeros) + 1;
459     gain32 *= gain32;
460     // check for overflow
461     while (AGC_MUL32((env[k] >> 12) + 1, gain32) >
462            WEBRTC_SPL_SHIFT_W32((int32_t)32767, 2 * (1 - zeros + 10))) {
463       // multiply by 253/256 ==> -0.1 dB
464       if (gains[k + 1] > 8388607) {
465         // Prevent wrap around
466         gains[k + 1] = (gains[k + 1] / 256) * 253;
467       } else {
468         gains[k + 1] = (gains[k + 1] * 253) / 256;
469       }
470       gain32 = (gains[k + 1] >> zeros) + 1;
471       gain32 *= gain32;
472     }
473   }
474   // gain reductions should be done 1 ms earlier than gain increases
475   for (k = 1; k < 10; k++) {
476     if (gains[k] > gains[k + 1]) {
477       gains[k] = gains[k + 1];
478     }
479   }
480   // save start gain for next frame
481   stt->gain = gains[10];
482 
483   return 0;
484 }
485 
WebRtcAgc_ApplyDigitalGains(const int32_t gains[11],size_t num_bands,uint32_t FS,const int16_t * const * in_near,int16_t * const * out)486 int32_t WebRtcAgc_ApplyDigitalGains(const int32_t gains[11],
487                                     size_t num_bands,
488                                     uint32_t FS,
489                                     const int16_t* const* in_near,
490                                     int16_t* const* out) {
491   // Apply gain
492   // handle first sub frame separately
493   size_t L;
494   int16_t L2;  // samples/subframe
495 
496   // determine number of samples per ms
497   if (FS == 8000) {
498     L = 8;
499     L2 = 3;
500   } else if (FS == 16000 || FS == 32000 || FS == 48000) {
501     L = 16;
502     L2 = 4;
503   } else {
504     return -1;
505   }
506 
507   for (size_t i = 0; i < num_bands; ++i) {
508     if (in_near[i] != out[i]) {
509       // Only needed if they don't already point to the same place.
510       memcpy(out[i], in_near[i], 10 * L * sizeof(in_near[i][0]));
511     }
512   }
513 
514   // iterate over samples
515   int32_t delta = (gains[1] - gains[0]) * (1 << (4 - L2));
516   int32_t gain32 = gains[0] * (1 << 4);
517   for (size_t n = 0; n < L; n++) {
518     for (size_t i = 0; i < num_bands; ++i) {
519       int32_t out_tmp = (int64_t)out[i][n] * ((gain32 + 127) >> 7) >> 16;
520       if (out_tmp > 4095) {
521         out[i][n] = (int16_t)32767;
522       } else if (out_tmp < -4096) {
523         out[i][n] = (int16_t)-32768;
524       } else {
525         int32_t tmp32 = ((int64_t)out[i][n] * (gain32 >> 4)) >> 16;
526         out[i][n] = (int16_t)tmp32;
527       }
528     }
529 
530     gain32 += delta;
531   }
532   // iterate over subframes
533   for (int k = 1; k < 10; k++) {
534     delta = (gains[k + 1] - gains[k]) * (1 << (4 - L2));
535     gain32 = gains[k] * (1 << 4);
536     // iterate over samples
537     for (size_t n = 0; n < L; n++) {
538       for (size_t i = 0; i < num_bands; ++i) {
539         int64_t tmp64 = ((int64_t)(out[i][k * L + n])) * (gain32 >> 4);
540         tmp64 = tmp64 >> 16;
541         if (tmp64 > 32767) {
542           out[i][k * L + n] = 32767;
543         } else if (tmp64 < -32768) {
544           out[i][k * L + n] = -32768;
545         } else {
546           out[i][k * L + n] = (int16_t)(tmp64);
547         }
548       }
549       gain32 += delta;
550     }
551   }
552   return 0;
553 }
554 
WebRtcAgc_InitVad(AgcVad * state)555 void WebRtcAgc_InitVad(AgcVad* state) {
556   int16_t k;
557 
558   state->HPstate = 0;   // state of high pass filter
559   state->logRatio = 0;  // log( P(active) / P(inactive) )
560   // average input level (Q10)
561   state->meanLongTerm = 15 << 10;
562 
563   // variance of input level (Q8)
564   state->varianceLongTerm = 500 << 8;
565 
566   state->stdLongTerm = 0;  // standard deviation of input level in dB
567   // short-term average input level (Q10)
568   state->meanShortTerm = 15 << 10;
569 
570   // short-term variance of input level (Q8)
571   state->varianceShortTerm = 500 << 8;
572 
573   state->stdShortTerm =
574       0;               // short-term standard deviation of input level in dB
575   state->counter = 3;  // counts updates
576   for (k = 0; k < 8; k++) {
577     // downsampling filter
578     state->downState[k] = 0;
579   }
580 }
581 
WebRtcAgc_ProcessVad(AgcVad * state,const int16_t * in,size_t nrSamples)582 int16_t WebRtcAgc_ProcessVad(AgcVad* state,       // (i) VAD state
583                              const int16_t* in,   // (i) Speech signal
584                              size_t nrSamples) {  // (i) number of samples
585   uint32_t nrg;
586   int32_t out, tmp32, tmp32b;
587   uint16_t tmpU16;
588   int16_t k, subfr, tmp16;
589   int16_t buf1[8];
590   int16_t buf2[4];
591   int16_t HPstate;
592   int16_t zeros, dB;
593   int64_t tmp64;
594 
595   // process in 10 sub frames of 1 ms (to save on memory)
596   nrg = 0;
597   HPstate = state->HPstate;
598   for (subfr = 0; subfr < 10; subfr++) {
599     // downsample to 4 kHz
600     if (nrSamples == 160) {
601       for (k = 0; k < 8; k++) {
602         tmp32 = (int32_t)in[2 * k] + (int32_t)in[2 * k + 1];
603         tmp32 >>= 1;
604         buf1[k] = (int16_t)tmp32;
605       }
606       in += 16;
607 
608       WebRtcSpl_DownsampleBy2(buf1, 8, buf2, state->downState);
609     } else {
610       WebRtcSpl_DownsampleBy2(in, 8, buf2, state->downState);
611       in += 8;
612     }
613 
614     // high pass filter and compute energy
615     for (k = 0; k < 4; k++) {
616       out = buf2[k] + HPstate;
617       tmp32 = 600 * out;
618       HPstate = (int16_t)((tmp32 >> 10) - buf2[k]);
619 
620       // Add 'out * out / 2**6' to 'nrg' in a non-overflowing
621       // way. Guaranteed to work as long as 'out * out / 2**6' fits in
622       // an int32_t.
623       nrg += out * (out / (1 << 6));
624       nrg += out * (out % (1 << 6)) / (1 << 6);
625     }
626   }
627   state->HPstate = HPstate;
628 
629   // find number of leading zeros
630   if (!(0xFFFF0000 & nrg)) {
631     zeros = 16;
632   } else {
633     zeros = 0;
634   }
635   if (!(0xFF000000 & (nrg << zeros))) {
636     zeros += 8;
637   }
638   if (!(0xF0000000 & (nrg << zeros))) {
639     zeros += 4;
640   }
641   if (!(0xC0000000 & (nrg << zeros))) {
642     zeros += 2;
643   }
644   if (!(0x80000000 & (nrg << zeros))) {
645     zeros += 1;
646   }
647 
648   // energy level (range {-32..30}) (Q10)
649   dB = (15 - zeros) * (1 << 11);
650 
651   // Update statistics
652 
653   if (state->counter < kAvgDecayTime) {
654     // decay time = AvgDecTime * 10 ms
655     state->counter++;
656   }
657 
658   // update short-term estimate of mean energy level (Q10)
659   tmp32 = state->meanShortTerm * 15 + dB;
660   state->meanShortTerm = (int16_t)(tmp32 >> 4);
661 
662   // update short-term estimate of variance in energy level (Q8)
663   tmp32 = (dB * dB) >> 12;
664   tmp32 += state->varianceShortTerm * 15;
665   state->varianceShortTerm = tmp32 / 16;
666 
667   // update short-term estimate of standard deviation in energy level (Q10)
668   tmp32 = state->meanShortTerm * state->meanShortTerm;
669   tmp32 = (state->varianceShortTerm << 12) - tmp32;
670   state->stdShortTerm = (int16_t)WebRtcSpl_Sqrt(tmp32);
671 
672   // update long-term estimate of mean energy level (Q10)
673   tmp32 = state->meanLongTerm * state->counter + dB;
674   state->meanLongTerm =
675       WebRtcSpl_DivW32W16ResW16(tmp32, WebRtcSpl_AddSatW16(state->counter, 1));
676 
677   // update long-term estimate of variance in energy level (Q8)
678   tmp32 = (dB * dB) >> 12;
679   tmp32 += state->varianceLongTerm * state->counter;
680   state->varianceLongTerm =
681       WebRtcSpl_DivW32W16(tmp32, WebRtcSpl_AddSatW16(state->counter, 1));
682 
683   // update long-term estimate of standard deviation in energy level (Q10)
684   tmp32 = state->meanLongTerm * state->meanLongTerm;
685   tmp32 = (state->varianceLongTerm << 12) - tmp32;
686   state->stdLongTerm = (int16_t)WebRtcSpl_Sqrt(tmp32);
687 
688   // update voice activity measure (Q10)
689   tmp16 = 3 << 12;
690   // TODO(bjornv): (dB - state->meanLongTerm) can overflow, e.g., in
691   // ApmTest.Process unit test. Previously the macro WEBRTC_SPL_MUL_16_16()
692   // was used, which did an intermediate cast to (int16_t), hence losing
693   // significant bits. This cause logRatio to max out positive, rather than
694   // negative. This is a bug, but has very little significance.
695   tmp32 = tmp16 * (int16_t)(dB - state->meanLongTerm);
696   tmp32 = WebRtcSpl_DivW32W16(tmp32, state->stdLongTerm);
697   tmpU16 = (13 << 12);
698   tmp32b = WEBRTC_SPL_MUL_16_U16(state->logRatio, tmpU16);
699   tmp64 = tmp32;
700   tmp64 += tmp32b >> 10;
701   tmp64 >>= 6;
702 
703   // limit
704   if (tmp64 > 2048) {
705     tmp64 = 2048;
706   } else if (tmp64 < -2048) {
707     tmp64 = -2048;
708   }
709   state->logRatio = (int16_t)tmp64;
710 
711   return state->logRatio;  // Q10
712 }
713 
714 }  // namespace webrtc
715