Moving FFT on farend signal to where it is used in AEC (bit exact).

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 823 matching lines @@
   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++) {
     freq_data[0][i] = time_data[2 * i];
     freq_data[1][i] = time_data[2 * i + 1];
   }
 }
 
 static int MoveFarReadPtrWithoutSystemDelayUpdate(AecCore* self, int elements) {

[peah-webrtc, 2015/12/16 12:52:22]

I think that since this is now a one-liner, we should remove the function and
instead call WebRtc_MoveReadPtr directly. This will increase code readability.

[minyue-webrtc, 2015/12/16 14:47:17]

yes, certainly.

-  WebRtc_MoveReadPtr(self->far_buf_windowed, elements);
-#ifdef WEBRTC_AEC_DEBUG_DUMP
-  WebRtc_MoveReadPtr(self->far_time_buf, elements);
-#endif
-  return WebRtc_MoveReadPtr(self->far_buf, elements);
+  return WebRtc_MoveReadPtr(self->far_time_buf, elements);
 }
 
 static int SignalBasedDelayCorrection(AecCore* self) {
   int delay_correction = 0;
   int last_delay = -2;
   assert(self != NULL);
 #if !defined(WEBRTC_ANDROID)
   // On desktops, turn on correction after |kDelayCorrectionStart| frames.  This
   // is to let the delay estimation get a chance to converge.  Also, if the
   // playout audio volume is low (or even muted) the delay estimation can return
@@ skipping 21 matching lines @@
            self->delay_quality_threshold)) {
     int delay = last_delay - WebRtc_lookahead(self->delay_estimator);
     // Allow for a slack in the actual delay, defined by a |lower_bound| and an
     // |upper_bound|.  The adaptive echo cancellation filter is currently
     // |num_partitions| (of 64 samples) long.  If the delay estimate is negative
     // or at least 3/4 of the filter length we open up for correction.
     const int lower_bound = 0;
     const int upper_bound = self->num_partitions * 3 / 4;
     const int do_correction = delay <= lower_bound || delay > upper_bound;
     if (do_correction == 1) {
-      int available_read = (int)WebRtc_available_read(self->far_buf);
+      int available_read = (int)WebRtc_available_read(self->far_time_buf);
       // With |shift_offset| we gradually rely on the delay estimates.  For
       // positive delays we reduce the correction by |shift_offset| to lower the
       // risk of pushing the AEC into a non causal state.  For negative delays
       // we rely on the values up to a rounding error, hence compensate by 1
       // element to make sure to push the delay into the causal region.
       delay_correction = -delay;
       delay_correction += delay > self->shift_offset ? self->shift_offset : 1;
       self->shift_offset--;
       self->shift_offset = (self->shift_offset <= 1 ? 1 : self->shift_offset);
       if (delay_correction > available_read - self->mult - 1) {
@@ skipping 94 matching lines @@
   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 farend[PART_LEN2],
                             float* echo_subtractor_output,
                             float* output,
                             float* const* outputH) {
   float efw[2][PART_LEN1];
   float xfw[2][PART_LEN1];
   float dfw[2][PART_LEN1];
   complex_t comfortNoiseHband[PART_LEN1];
   float fft[PART_LEN2];
   float scale, dtmp;
   float nlpGainHband;
@@ skipping 33 matching lines @@
   // Windowed echo suppressor output ffts.
   WindowData(fft, aec->eBuf);
   aec_rdft_forward_128(fft);
   StoreAsComplex(fft, efw);
 
   aec->delayEstCtr++;
   if (aec->delayEstCtr == delayEstInterval) {
     aec->delayEstCtr = 0;
   }
 
-  // We should always have at least one element stored in |far_buf|.
-  assert(WebRtc_available_read(aec->far_buf_windowed) > 0);
   // NLP
-  WebRtc_ReadBuffer(aec->far_buf_windowed, (void**)&xfw_ptr, &xfw[0][0], 1);
 
-  // TODO(bjornv): Investigate if we can reuse |far_buf_windowed| instead of
-  // |xfwBuf|.
+  // Convert far-end partition to the frequency domain with windowing.
+  WindowData(fft, farend);
+  Fft(fft, xfw);
+  xfw_ptr = &xfw[0][0];
+
   // Buffer far.
   memcpy(aec->xfwBuf, xfw_ptr, sizeof(float) * 2 * PART_LEN1);
 
   if (aec->delayEstCtr == 0)
     aec->delayIdx = WebRtcAec_PartitionDelay(aec);
 
   // Use delayed far.
   memcpy(xfw,
          aec->xfwBuf + aec->delayIdx * PART_LEN1,
          sizeof(xfw[0][0]) * 2 * PART_LEN1);
@@ skipping 202 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* nearend_ptr = NULL;
+  float farend[PART_LEN2];
+  float* farend_ptr = NULL;
   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];
   }
 
   // Concatenate old and new nearend blocks.
   for (i = 0; i < aec->num_bands - 1; ++i) {
     WebRtc_ReadBuffer(aec->nearFrBufH[i],
                       (void**)&nearend_ptr,
                       nearend,
                       PART_LEN);
     memcpy(aec->dBufH[i] + PART_LEN, nearend_ptr, sizeof(nearend));
   }
   WebRtc_ReadBuffer(aec->nearFrBuf, (void**)&nearend_ptr, nearend, PART_LEN);
   memcpy(aec->dBuf + PART_LEN, nearend_ptr, sizeof(nearend));
 
-  // ---------- Ooura fft ----------
+  // We should always have at least one element stored in |far_buf|.
+  assert(WebRtc_available_read(aec->far_time_buf) > 0);
+  WebRtc_ReadBuffer(aec->far_time_buf, (void**)&farend_ptr, farend, 1);
 
 #ifdef WEBRTC_AEC_DEBUG_DUMP
   {
-    float farend[PART_LEN];
-    float* farend_ptr = NULL;
-    WebRtc_ReadBuffer(aec->far_time_buf, (void**)&farend_ptr, farend, 1);
-    RTC_AEC_DEBUG_WAV_WRITE(aec->farFile, farend_ptr, PART_LEN);
+    // TODO(minyue): |farend_ptr| starts from buffered samples. This will be
+    // modified when |aec->far_time_buf| is revised.
+    RTC_AEC_DEBUG_WAV_WRITE(aec->farFile, &farend_ptr[PART_LEN], PART_LEN);
+
     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);
+  // Convert far-end signal to the frequency domain.
+  memcpy(fft, farend_ptr, sizeof(float) * PART_LEN2);
+  Fft(fft, xf);
+  xf_ptr = &xf[0][0];
 
   // Near fft
   memcpy(fft, aec->dBuf, sizeof(float) * PART_LEN2);
   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] =
@@ skipping 79 matching lines @@
                   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.
-  EchoSuppression(aec, echo_subtractor_output, output, outputH_ptr);
+  EchoSuppression(aec, farend_ptr, echo_subtractor_output, 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.
   WebRtc_WriteBuffer(aec->outFrBuf, output, PART_LEN);
@@ skipping 33 matching lines @@
     }
     aec->outFrBufH[i] = WebRtc_CreateBuffer(FRAME_LEN + PART_LEN,
                                             sizeof(float));
     if (!aec->outFrBufH[i]) {
       WebRtcAec_FreeAec(aec);
       return NULL;
     }
   }
 
   // Create far-end buffers.
-  aec->far_buf =
-      WebRtc_CreateBuffer(kBufSizePartitions, sizeof(float) * 2 * PART_LEN1);
-  if (!aec->far_buf) {
-    WebRtcAec_FreeAec(aec);
-    return NULL;
-  }
-  aec->far_buf_windowed =
-      WebRtc_CreateBuffer(kBufSizePartitions, sizeof(float) * 2 * PART_LEN1);
-  if (!aec->far_buf_windowed) {
-    WebRtcAec_FreeAec(aec);
-    return NULL;
-  }
-#ifdef WEBRTC_AEC_DEBUG_DUMP
-  aec->instance_index = webrtc_aec_instance_count;
+  // For bit exactness with legacy code, each element in |far_time_buf| is
+  // supposed to contain |PART_LEN2| samples with an overlap of |PART_LEN|
+  // samples from the last frame.
+  // TODO(minyue): reduce |far_time_buf| to non-overlapped |PART_LEN| samples.
   aec->far_time_buf =
-      WebRtc_CreateBuffer(kBufSizePartitions, sizeof(float) * PART_LEN);
+      WebRtc_CreateBuffer(kBufSizePartitions, sizeof(float) * PART_LEN2);
   if (!aec->far_time_buf) {
     WebRtcAec_FreeAec(aec);
     return NULL;
   }
+
+#ifdef WEBRTC_AEC_DEBUG_DUMP
+  aec->instance_index = webrtc_aec_instance_count;
+
   aec->farFile = aec->nearFile = aec->outFile = aec->outLinearFile = NULL;
   aec->debug_dump_count = 0;
 #endif
   aec->delay_estimator_farend =
       WebRtc_CreateDelayEstimatorFarend(PART_LEN1, kHistorySizeBlocks);
   if (aec->delay_estimator_farend == NULL) {
     WebRtcAec_FreeAec(aec);
     return NULL;
   }
   // We create the delay_estimator with the same amount of maximum lookahead as
@@ skipping 57 matching lines @@
   }
 
   WebRtc_FreeBuffer(aec->nearFrBuf);
   WebRtc_FreeBuffer(aec->outFrBuf);
 
   for (i = 0; i < NUM_HIGH_BANDS_MAX; ++i) {
     WebRtc_FreeBuffer(aec->nearFrBufH[i]);
     WebRtc_FreeBuffer(aec->outFrBufH[i]);
   }
 
-  WebRtc_FreeBuffer(aec->far_buf);
-  WebRtc_FreeBuffer(aec->far_buf_windowed);
-#ifdef WEBRTC_AEC_DEBUG_DUMP
   WebRtc_FreeBuffer(aec->far_time_buf);
-#endif
+
   RTC_AEC_DEBUG_WAV_CLOSE(aec->farFile);
   RTC_AEC_DEBUG_WAV_CLOSE(aec->nearFile);
   RTC_AEC_DEBUG_WAV_CLOSE(aec->outFile);
   RTC_AEC_DEBUG_WAV_CLOSE(aec->outLinearFile);
   RTC_AEC_DEBUG_RAW_CLOSE(aec->e_fft_file);
 
   WebRtc_FreeDelayEstimator(aec->delay_estimator);
   WebRtc_FreeDelayEstimatorFarend(aec->delay_estimator_farend);
 
   free(aec);
@@ skipping 15 matching lines @@
   }
 
   WebRtc_InitBuffer(aec->nearFrBuf);
   WebRtc_InitBuffer(aec->outFrBuf);
   for (i = 0; i < NUM_HIGH_BANDS_MAX; ++i) {
     WebRtc_InitBuffer(aec->nearFrBufH[i]);
     WebRtc_InitBuffer(aec->outFrBufH[i]);
   }
 
   // Initialize far-end buffers.
-  WebRtc_InitBuffer(aec->far_buf);
-  WebRtc_InitBuffer(aec->far_buf_windowed);
+  WebRtc_InitBuffer(aec->far_time_buf);
+
 #ifdef WEBRTC_AEC_DEBUG_DUMP
-  WebRtc_InitBuffer(aec->far_time_buf);
   {
     int process_rate = sampFreq > 16000 ? 16000 : sampFreq;
     RTC_AEC_DEBUG_WAV_REOPEN("aec_far", aec->instance_index,
                              aec->debug_dump_count, process_rate,
                              &aec->farFile );
     RTC_AEC_DEBUG_WAV_REOPEN("aec_near", aec->instance_index,
                              aec->debug_dump_count, process_rate,
                              &aec->nearFile);
     RTC_AEC_DEBUG_WAV_REOPEN("aec_out", aec->instance_index,
                              aec->debug_dump_count, process_rate,
@@ skipping 123 matching lines @@
 
   aec->extreme_filter_divergence = 0;
 
   // Metrics disabled by default
   aec->metricsMode = 0;
   InitMetrics(aec);
 
   return 0;
 }
 
+
+// For bit exactness with a legacy code, |farend| is supposed to contain
+// |PART_LEN2| samples with an overlap of |PART_LEN| samples from the last
+// frame.
+// TODO(minyue): reduce |farend| to non-overlapped |PART_LEN| samples.
 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) {
+  if (WebRtc_available_write(aec->far_time_buf) < 1) {
     WebRtcAec_MoveFarReadPtr(aec, 1);
   }
-  // Convert far-end partition to the frequency domain without windowing.
-  memcpy(fft, farend, sizeof(float) * PART_LEN2);
-  Fft(fft, xf);
-  WebRtc_WriteBuffer(aec->far_buf, &xf[0][0], 1);
 
-  // Convert far-end partition to the frequency domain with windowing.
-  WindowData(fft, farend);
-  Fft(fft, xf);
-  WebRtc_WriteBuffer(aec->far_buf_windowed, &xf[0][0], 1);
+  WebRtc_WriteBuffer(aec->far_time_buf, farend, 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,
                              const float* const* nearend,
@@ skipping 64 matching lines @@
       // rounding, like -16.
       int move_elements = (aec->knownDelay - knownDelay - 32) / PART_LEN;
       int moved_elements =
           MoveFarReadPtrWithoutSystemDelayUpdate(aec, move_elements);
       aec->knownDelay -= moved_elements * PART_LEN;
     } else {
       // 2 b) Apply signal based delay correction.
       int move_elements = SignalBasedDelayCorrection(aec);
       int moved_elements =
           MoveFarReadPtrWithoutSystemDelayUpdate(aec, move_elements);
-      int far_near_buffer_diff = WebRtc_available_read(aec->far_buf) -
+      int far_near_buffer_diff = WebRtc_available_read(aec->far_time_buf) -
           WebRtc_available_read(aec->nearFrBuf) / PART_LEN;
       WebRtc_SoftResetDelayEstimator(aec->delay_estimator, moved_elements);
       WebRtc_SoftResetDelayEstimatorFarend(aec->delay_estimator_farend,
                                            moved_elements);
       aec->signal_delay_correction += moved_elements;
       // If we rely on reported system delay values only, a buffer underrun here
       // can never occur since we've taken care of that in 1) above.  Here, we
       // apply signal based delay correction and can therefore end up with
       // buffer underruns since the delay estimation can be wrong.  We therefore
       // stuff the buffer with enough elements if needed.
@@ skipping 58 matching lines @@
                             Stats* erle,
                             Stats* a_nlp) {
   assert(erl != NULL);
   assert(erle != NULL);
   assert(a_nlp != NULL);
   *erl = self->erl;
   *erle = self->erle;
   *a_nlp = self->aNlp;
 }
 
-#ifdef WEBRTC_AEC_DEBUG_DUMP
-void* WebRtcAec_far_time_buf(AecCore* self) { return self->far_time_buf; }
-#endif
-
 void WebRtcAec_SetConfigCore(AecCore* self,
                              int nlp_mode,
                              int metrics_mode,
                              int delay_logging) {
   assert(nlp_mode >= 0 && nlp_mode < 3);
   self->nlp_mode = nlp_mode;
   self->metricsMode = metrics_mode;
   if (self->metricsMode) {
     InitMetrics(self);
   }
@@ skipping 23 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;
 }