Ducking fix #3: Removed the state as an input to the FilterAdaptation function

webrtc/modules/audio_processing/aec/aec_core.c

 /*
  *  Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 133 matching lines @@
   return aRe * bIm + aIm * bRe;
 }
 
 static int CmpFloat(const void* a, const void* b) {
   const float* da = (const float*)a;
   const float* db = (const float*)b;
 
   return (*da > *db) - (*da < *db);
 }
 
-static void FilterFar(int num_partitions,
-                      int xfBufBlockPos,
-                      float xfBuf[2][kExtendedNumPartitions * PART_LEN1],
-                      float wfBuf[2][kExtendedNumPartitions * PART_LEN1],
-                      float yf[2][PART_LEN1]) {
+static void FilterFar(
+    int num_partitions,
+    int x_fft_buf_block_pos,
+    const float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
+    const float h_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
+    float y_fft[2][PART_LEN1]) {
   int i;
   for (i = 0; i < num_partitions; i++) {
     int j;
-    int xPos = (i + xfBufBlockPos) * PART_LEN1;
+    int xPos = (i + x_fft_buf_block_pos) * PART_LEN1;
     int pos = i * PART_LEN1;
     // Check for wrap
-    if (i + xfBufBlockPos >= num_partitions) {
+    if (i + x_fft_buf_block_pos >= num_partitions) {
       xPos -= num_partitions * (PART_LEN1);
     }
 
     for (j = 0; j < PART_LEN1; j++) {
-      yf[0][j] += MulRe(xfBuf[0][xPos + j],
-                        xfBuf[1][xPos + j],
-                        wfBuf[0][pos + j],
-                        wfBuf[1][pos + j]);
-      yf[1][j] += MulIm(xfBuf[0][xPos + j],
-                        xfBuf[1][xPos + j],
-                        wfBuf[0][pos + j],
-                        wfBuf[1][pos + j]);
+      y_fft[0][j] += MulRe(x_fft_buf[0][xPos + j],
+                           x_fft_buf[1][xPos + j],
+                           h_fft_buf[0][pos + j],
+                           h_fft_buf[1][pos + j]);
+      y_fft[1][j] += MulIm(x_fft_buf[0][xPos + j],
+                           x_fft_buf[1][xPos + j],
+                           h_fft_buf[0][pos + j],
+                           h_fft_buf[1][pos + j]);
     }
   }
 }
 
 static void ScaleErrorSignal(int extended_filter_enabled,
                              float normal_mu,
                              float normal_error_threshold,
-                             float *xPow,
+                             const float xPow[PART_LEN1],
                              float ef[2][PART_LEN1]) {
   const float mu = extended_filter_enabled ? kExtendedMu : normal_mu;
   const float error_threshold = extended_filter_enabled
                                     ? kExtendedErrorThreshold
                                     : normal_error_threshold;
   int i;
   float abs_ef;
   for (i = 0; i < (PART_LEN1); i++) {
     ef[0][i] /= (xPow[i] + 1e-10f);
     ef[1][i] /= (xPow[i] + 1e-10f);
     abs_ef = sqrtf(ef[0][i] * ef[0][i] + ef[1][i] * ef[1][i]);
 
     if (abs_ef > error_threshold) {
       abs_ef = error_threshold / (abs_ef + 1e-10f);
       ef[0][i] *= abs_ef;
       ef[1][i] *= abs_ef;
     }
 
     // Stepsize factor
     ef[0][i] *= mu;
     ef[1][i] *= mu;
   }
 }
 
-// Time-unconstrined filter adaptation.
-// TODO(andrew): consider for a low-complexity mode.
-// static void FilterAdaptationUnconstrained(AecCore* aec, float *fft,
-//                                          float ef[2][PART_LEN1]) {
-//  int i, j;
-//  for (i = 0; i < aec->num_partitions; i++) {
-//    int xPos = (i + aec->xfBufBlockPos)*(PART_LEN1);
-//    int pos;
-//    // Check for wrap
-//    if (i + aec->xfBufBlockPos >= aec->num_partitions) {
-//      xPos -= aec->num_partitions * PART_LEN1;
-//    }
-//
-//    pos = i * PART_LEN1;
-//
-//    for (j = 0; j < PART_LEN1; j++) {
-//      aec->wfBuf[0][pos + j] += MulRe(aec->xfBuf[0][xPos + j],
-//                                      -aec->xfBuf[1][xPos + j],
-//                                      ef[0][j], ef[1][j]);
-//      aec->wfBuf[1][pos + j] += MulIm(aec->xfBuf[0][xPos + j],
-//                                      -aec->xfBuf[1][xPos + j],
-//                                      ef[0][j], ef[1][j]);
-//    }
-//  }
-//}
 
-static void FilterAdaptation(AecCore* aec, float* fft, float ef[2][PART_LEN1]) {
+static void FilterAdaptation(
+    int num_partitions,
+    int x_fft_buf_block_pos,
+    const float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
+    const float e_fft[2][PART_LEN1],
+    float h_fft_buf[2][kExtendedNumPartitions * PART_LEN1]) {
   int i, j;
-  for (i = 0; i < aec->num_partitions; i++) {
-    int xPos = (i + aec->xfBufBlockPos) * (PART_LEN1);
+  float fft[PART_LEN2];
+  for (i = 0; i < num_partitions; i++) {
+    int xPos = (i + x_fft_buf_block_pos) * (PART_LEN1);
     int pos;
     // Check for wrap
-    if (i + aec->xfBufBlockPos >= aec->num_partitions) {
-      xPos -= aec->num_partitions * PART_LEN1;
+    if (i + x_fft_buf_block_pos >= num_partitions) {
+      xPos -= num_partitions * PART_LEN1;
     }
 
     pos = i * PART_LEN1;
 
     for (j = 0; j < PART_LEN; j++) {
 
-      fft[2 * j] = MulRe(aec->xfBuf[0][xPos + j],
-                         -aec->xfBuf[1][xPos + j],
-                         ef[0][j],
-                         ef[1][j]);
-      fft[2 * j + 1] = MulIm(aec->xfBuf[0][xPos + j],
-                             -aec->xfBuf[1][xPos + j],
-                             ef[0][j],
-                             ef[1][j]);
+      fft[2 * j] = MulRe(x_fft_buf[0][xPos + j],
+                         -x_fft_buf[1][xPos + j],
+                         e_fft[0][j],
+                         e_fft[1][j]);
+      fft[2 * j + 1] = MulIm(x_fft_buf[0][xPos + j],
+                             -x_fft_buf[1][xPos + j],
+                             e_fft[0][j],
+                             e_fft[1][j]);
     }
-    fft[1] = MulRe(aec->xfBuf[0][xPos + PART_LEN],
-                   -aec->xfBuf[1][xPos + PART_LEN],
-                   ef[0][PART_LEN],
-                   ef[1][PART_LEN]);
+    fft[1] = MulRe(x_fft_buf[0][xPos + PART_LEN],
+                   -x_fft_buf[1][xPos + PART_LEN],
+                   e_fft[0][PART_LEN],
+                   e_fft[1][PART_LEN]);
 
     aec_rdft_inverse_128(fft);
     memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN);
 
     // fft scaling
     {
       float scale = 2.0f / PART_LEN2;
       for (j = 0; j < PART_LEN; j++) {
         fft[j] *= scale;
       }
     }
     aec_rdft_forward_128(fft);
 
-    aec->wfBuf[0][pos] += fft[0];
-    aec->wfBuf[0][pos + PART_LEN] += fft[1];
+    h_fft_buf[0][pos] += fft[0];
+    h_fft_buf[0][pos + PART_LEN] += fft[1];
 
     for (j = 1; j < PART_LEN; j++) {
-      aec->wfBuf[0][pos + j] += fft[2 * j];
-      aec->wfBuf[1][pos + j] += fft[2 * j + 1];
+      h_fft_buf[0][pos + j] += fft[2 * j];
+      h_fft_buf[1][pos + j] += fft[2 * j + 1];
     }
   }
 }
 
 static void OverdriveAndSuppress(AecCore* aec,
                                  float hNl[PART_LEN1],
                                  const float hNlFb,
                                  float efw[2][PART_LEN1]) {
   int i;
   for (i = 0; i < PART_LEN1; i++) {
@@ skipping 546 matching lines @@
         self->num_delay_values;
   }
 
   // Reset histogram.
   memset(self->delay_histogram, 0, sizeof(self->delay_histogram));
   self->num_delay_values = 0;
 
   return;
 }
 
-static void FrequencyToTime(float freq_data[2][PART_LEN1],
-                            float time_data[PART_LEN2]) {
+static void InverseFft(const float freq_data[2][PART_LEN1],
+                       float time_data[PART_LEN2]) {
+  const float scale = 1.0f / PART_LEN2;
   time_data[0] = freq_data[0][0];
   time_data[1] = freq_data[0][PART_LEN];
   for (int i = 1; i < PART_LEN; i++) {
     time_data[2 * i] = freq_data[0][i];
     time_data[2 * i + 1] = freq_data[1][i];
   }
   aec_rdft_inverse_128(time_data);
+
+  for (int i = 0; i < PART_LEN2; i++) {
+    time_data[i] *= scale;

[ivoc, 2015/11/26 12:46:44]

I think it is more efficient to merge this loop with the previous one, i.e.:
time_data[2*i] = freq_data[0][i] * scale;

[peah-webrtc, 2015/11/27 04:58:36]

I'll do that. Before I did not to that for fear of running into some scale
problem with the FFT lib, but this should be verifiable in the bitexactness
tests, so I'll try to add that. 

Done.

+  }
 }
 
 
-static void TimeToFrequency(float time_data[PART_LEN2],
-                            float freq_data[2][PART_LEN1],
-                            int window) {
-  int i = 0;
-
-  // TODO(bjornv): Should we have a different function/wrapper for windowed FFT?
-  if (window) {
-    for (i = 0; i < PART_LEN; i++) {
-      time_data[i] *= WebRtcAec_sqrtHanning[i];
-      time_data[PART_LEN + i] *= WebRtcAec_sqrtHanning[PART_LEN - i];
-    }
-  }
+static void Fft(float time_data[PART_LEN2],
+                float freq_data[2][PART_LEN1]) {
+  // Make local copy of the time_data to avoid changing it.
+  float time_data_copy[PART_LEN2];
+  memcpy(time_data_copy, time_data, sizeof(float) * PART_LEN2);

[hlundin-webrtc, 2015/11/24 13:51:44]

I'm not sure I like yet another memcpy of all audio data...

[peah-webrtc, 2015/11/26 05:55:17]

I'll remote it.
Done.

[hlundin-webrtc, 2015/11/26 08:41:44]

Great. Btw: did you even use it?

[peah-webrtc, 2015/11/27 04:58:36]

No. So safe removal it is! :-)

 
   aec_rdft_forward_128(time_data);
-  // Reorder.
+
+  // Reorder fft output data.
   freq_data[1][0] = 0;
   freq_data[1][PART_LEN] = 0;
   freq_data[0][0] = time_data[0];
   freq_data[0][PART_LEN] = time_data[1];
-  for (i = 1; i < PART_LEN; i++) {
+  for (int i = 1; i < PART_LEN; i++) {
     freq_data[0][i] = time_data[2 * i];
     freq_data[1][i] = time_data[2 * i + 1];
   }
 }
 
 static int MoveFarReadPtrWithoutSystemDelayUpdate(AecCore* self, int elements) {
   WebRtc_MoveReadPtr(self->far_buf_windowed, elements);
 #ifdef WEBRTC_AEC_DEBUG_DUMP
   WebRtc_MoveReadPtr(self->far_time_buf, elements);
 #endif
@@ skipping 65 matching lines @@
     float delay_quality = WebRtc_last_delay_quality(self->delay_estimator);
     delay_quality = (delay_quality > kDelayQualityThresholdMax ?
         kDelayQualityThresholdMax : delay_quality);
     self->delay_quality_threshold =
         (delay_quality > self->delay_quality_threshold ? delay_quality :
             self->delay_quality_threshold);
   }
   return delay_correction;
 }
 
-static void EchoSubtraction(AecCore* aec,
-                            float* nearend_ptr) {
-  float yf[2][PART_LEN1];
-  float fft[PART_LEN2];
-  float y[PART_LEN];
+static void EchoSubtraction(
+    AecCore* aec,
+    int num_partitions,
+    int x_fft_buf_block_pos,
+    int metrics_mode,
+    int extended_filter_enabled,
+    float normal_mu,
+    float normal_error_threshold,
+    const float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
+    const float* const y,
+    const float x_pow[PART_LEN1],
+    float h_fft_buf[2][kExtendedNumPartitions * PART_LEN1],
+    PowerLevel* linout_level,
+    float echo_subtractor_output[PART_LEN]) {
+  float s_fft[2][PART_LEN1];
+  float e_extended[PART_LEN2];
+  float s_extended[PART_LEN2];
+  float *s;
   float e[PART_LEN];
-  float ef[2][PART_LEN1];
-  float scale;
+  float e_fft[2][PART_LEN1];
   int i;
-  memset(yf, 0, sizeof(yf));
+  memset(s_fft, 0, sizeof(s_fft));
 
-  // Produce echo estimate.
-  WebRtcAec_FilterFar(aec->num_partitions,
-                      aec->xfBufBlockPos,
-                      aec->xfBuf,
-                      aec->wfBuf,
-                      yf);
+  // Produce echo estimate s_fft.
+  WebRtcAec_FilterFar(num_partitions,
+                      x_fft_buf_block_pos,
+                      x_fft_buf,
+                      h_fft_buf,
+                      s_fft);
 
-  // Inverse fft to obtain echo estimate and error.
-  FrequencyToTime(yf, fft);
-
-  // Extract the output signal and compute the time-domain error
-  scale = 2.0f / PART_LEN2;
-  for (i = 0; i < PART_LEN; i++) {
-    y[i] = fft[PART_LEN + i] * scale;  // fft scaling
+  // Compute the time-domain echo estimate s.
+  InverseFft(s_fft, s_extended);
+  s = &s_extended[PART_LEN];
+  for (i = 0; i < PART_LEN; ++i) {
+    s[i] *= 2.0f;
   }
 
-  for (i = 0; i < PART_LEN; i++) {
-    e[i] = nearend_ptr[i] - y[i];
+  // Compute the time-domain echo prediction error.
+  for (i = 0; i < PART_LEN; ++i) {
+    e[i] = y[i] - s[i];
   }
 
+  // Compute the frequency domain echo prediction error.
+  memset(e_extended, 0, sizeof(float) * PART_LEN);
+  memcpy(e_extended + PART_LEN, e, sizeof(float) * PART_LEN);
+  Fft(e_extended, e_fft);
 
-  // Error fft
-  memcpy(aec->eBuf + PART_LEN, e, sizeof(float) * PART_LEN);
-  memset(fft, 0, sizeof(float) * PART_LEN);
-  memcpy(fft + PART_LEN, e, sizeof(float) * PART_LEN);
-  TimeToFrequency(fft, ef, 0);
 
   RTC_AEC_DEBUG_RAW_WRITE(aec->e_fft_file,
-                          &ef[0][0],
-                          sizeof(ef[0][0]) * PART_LEN1 * 2);
+                          &e_fft[0][0],
+                          sizeof(e_fft[0][0]) * PART_LEN1 * 2);
 
-  if (aec->metricsMode == 1) {
+  if (metrics_mode == 1) {
     // Note that the first PART_LEN samples in fft (before transformation) are
     // zero. Hence, the scaling by two in UpdateLevel() should not be
     // performed. That scaling is taken care of in UpdateMetrics() instead.
-    UpdateLevel(&aec->linoutlevel, ef);
+    UpdateLevel(linout_level, e_fft);
   }
 
   // Scale error signal inversely with far power.
-  WebRtcAec_ScaleErrorSignal(aec->extended_filter_enabled,
-                             aec->normal_mu,
-                             aec->normal_error_threshold,
-                             aec->xPow,
-                             ef);
-  WebRtcAec_FilterAdaptation(aec, fft, ef);
-
-
-  RTC_AEC_DEBUG_WAV_WRITE(aec->outLinearFile, e, PART_LEN);
+  WebRtcAec_ScaleErrorSignal(extended_filter_enabled,
+                             normal_mu,
+                             normal_error_threshold,
+                             x_pow,
+                             e_fft);
+  WebRtcAec_FilterAdaptation(num_partitions,
+                             x_fft_buf_block_pos,
+                             x_fft_buf,
+                             e_fft,
+                             h_fft_buf);
+  memcpy(echo_subtractor_output, e, sizeof(float) * PART_LEN);
 }
 
 
 static void EchoSuppression(AecCore* aec,
                             float* output,
                             float* const* outputH) {
   float efw[2][PART_LEN1], xfw[2][PART_LEN1];
   complex_t comfortNoiseHband[PART_LEN1];
   float fft[PART_LEN2];
   float scale, dtmp;
@@ skipping 236 matching lines @@
 
   const float gPow[2] = {0.9f, 0.1f};
 
   // Noise estimate constants.
   const int noiseInitBlocks = 500 * aec->mult;
   const float step = 0.1f;
   const float ramp = 1.0002f;
   const float gInitNoise[2] = {0.999f, 0.001f};
 
   float nearend[PART_LEN];
+  float echo_subtractor_output[PART_LEN];
   float* nearend_ptr = NULL;
   float output[PART_LEN];
   float outputH[NUM_HIGH_BANDS_MAX][PART_LEN];
   float* outputH_ptr[NUM_HIGH_BANDS_MAX];
   float* xf_ptr = NULL;
 
   for (i = 0; i < NUM_HIGH_BANDS_MAX; ++i) {
     outputH_ptr[i] = outputH[i];
   }
 
@@ skipping 19 matching lines @@
     RTC_AEC_DEBUG_WAV_WRITE(aec->nearFile, nearend_ptr, PART_LEN);
   }
 #endif
 
   // We should always have at least one element stored in |far_buf|.
   assert(WebRtc_available_read(aec->far_buf) > 0);
   WebRtc_ReadBuffer(aec->far_buf, (void**)&xf_ptr, &xf[0][0], 1);
 
   // Near fft
   memcpy(fft, aec->dBuf, sizeof(float) * PART_LEN2);
-  TimeToFrequency(fft, df, 0);
+  Fft(fft, df);
 
   // Power smoothing
   for (i = 0; i < PART_LEN1; i++) {
     far_spectrum = (xf_ptr[i] * xf_ptr[i]) +
                    (xf_ptr[PART_LEN1 + i] * xf_ptr[PART_LEN1 + i]);
     aec->xPow[i] =
         gPow[0] * aec->xPow[i] + gPow[1] * aec->num_partitions * far_spectrum;
     // Calculate absolute spectra
     abs_far_spectrum[i] = sqrtf(far_spectrum);
 
@@ skipping 58 matching lines @@
 
   // Buffer xf
   memcpy(aec->xfBuf[0] + aec->xfBufBlockPos * PART_LEN1,
          xf_ptr,
          sizeof(float) * PART_LEN1);
   memcpy(aec->xfBuf[1] + aec->xfBufBlockPos * PART_LEN1,
          &xf_ptr[PART_LEN1],
          sizeof(float) * PART_LEN1);
 
   // Perform echo subtraction.
-  EchoSubtraction(aec, nearend_ptr);
+  EchoSubtraction(aec,
+                  aec->num_partitions,
+                  aec->xfBufBlockPos,
+                  aec->metricsMode,
+                  aec->extended_filter_enabled,
+                  aec->normal_mu,
+                  aec->normal_error_threshold,
+                  aec->xfBuf,
+                  nearend_ptr,
+                  aec->xPow,
+                  aec->wfBuf,
+                  &aec->linoutlevel,
+                  echo_subtractor_output);
+
+  RTC_AEC_DEBUG_WAV_WRITE(aec->outLinearFile, echo_subtractor_output, PART_LEN);
 
   // Perform echo suppression.
+  memcpy(aec->eBuf + PART_LEN,
+         echo_subtractor_output,
+         sizeof(float) * PART_LEN);
   EchoSuppression(aec, output, outputH_ptr);
 
   if (aec->metricsMode == 1) {
     // Update power levels and echo metrics
     UpdateLevel(&aec->farlevel, (float(*)[PART_LEN1])xf_ptr);
     UpdateLevel(&aec->nearlevel, df);
     UpdateMetrics(aec);
   }
 
   // Store the output block.
@@ skipping 322 matching lines @@
 void WebRtcAec_BufferFarendPartition(AecCore* aec, const float* farend) {
   float fft[PART_LEN2];
   float xf[2][PART_LEN1];
 
   // Check if the buffer is full, and in that case flush the oldest data.
   if (WebRtc_available_write(aec->far_buf) < 1) {
     WebRtcAec_MoveFarReadPtr(aec, 1);
   }
   // Convert far-end partition to the frequency domain without windowing.
   memcpy(fft, farend, sizeof(float) * PART_LEN2);
-  TimeToFrequency(fft, xf, 0);
+  Fft(fft, xf);
   WebRtc_WriteBuffer(aec->far_buf, &xf[0][0], 1);
 
   // Convert far-end partition to the frequency domain with windowing.
-  memcpy(fft, farend, sizeof(float) * PART_LEN2);
-  TimeToFrequency(fft, xf, 1);
+  WindowData(fft, farend);
+  Fft(fft, xf);
   WebRtc_WriteBuffer(aec->far_buf_windowed, &xf[0][0], 1);
 }
 
 int WebRtcAec_MoveFarReadPtr(AecCore* aec, int elements) {
   int elements_moved = MoveFarReadPtrWithoutSystemDelayUpdate(aec, elements);
   aec->system_delay -= elements_moved * PART_LEN;
   return elements_moved;
 }
 
 void WebRtcAec_ProcessFrames(AecCore* aec,
@@ skipping 191 matching lines @@
 int WebRtcAec_extended_filter_enabled(AecCore* self) {
   return self->extended_filter_enabled;
 }
 
 int WebRtcAec_system_delay(AecCore* self) { return self->system_delay; }
 
 void WebRtcAec_SetSystemDelay(AecCore* self, int delay) {
   assert(delay >= 0);
   self->system_delay = delay;
 }