Avoiding overflow in cross correlation in NetEq.

webrtc/modules/audio_coding/neteq/time_stretch.cc

 /*
  *  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.
  */
 
 #include "webrtc/modules/audio_coding/neteq/time_stretch.h"
 
 #include <algorithm>  // min, max
 #include <memory>
 
 #include "webrtc/base/safe_conversions.h"
 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
 #include "webrtc/modules/audio_coding/neteq/background_noise.h"
+#include "webrtc/modules/audio_coding/neteq/cross_correlation.h"
 #include "webrtc/modules/audio_coding/neteq/dsp_helper.h"
 
 namespace webrtc {
 
 TimeStretch::ReturnCodes TimeStretch::Process(const int16_t* input,
                                               size_t input_len,
                                               bool fast_mode,
                                               AudioMultiVector* output,
                                               size_t* length_change_samples) {
   // Pre-calculate common multiplication with |fs_mult_|.
@@ skipping 122 matching lines @@
       break;
     case kNoStretch:
     case kError:
       *length_change_samples = 0;
       break;
   }
   return return_value;
 }
 
 void TimeStretch::AutoCorrelation() {
-  // Set scaling factor for cross correlation to protect against overflow.
-  int scaling = kLogCorrelationLen - WebRtcSpl_NormW32(
-      max_input_value_ * max_input_value_);
-  scaling = std::max(0, scaling);
-
   // Calculate correlation from lag kMinLag to lag kMaxLag in 4 kHz domain.
   int32_t auto_corr[kCorrelationLen];
-  WebRtcSpl_CrossCorrelation(auto_corr, &downsampled_input_[kMaxLag],
-                             &downsampled_input_[kMaxLag - kMinLag],
-                             kCorrelationLen, kMaxLag - kMinLag, scaling, -1);
+  CrossCorrelationWithAutoShift(
+      auto_corr, &downsampled_input_[kMaxLag],
+      &downsampled_input_[kMaxLag - kMinLag], kCorrelationLen,
+      kMaxLag - kMinLag, -1);
 
   // Normalize correlation to 14 bits and write to |auto_correlation_|.
   int32_t max_corr = WebRtcSpl_MaxAbsValueW32(auto_corr, kCorrelationLen);
-  scaling = std::max(0, 17 - WebRtcSpl_NormW32(max_corr));
+  int scaling = std::max(0, 17 - WebRtcSpl_NormW32(max_corr));
   WebRtcSpl_VectorBitShiftW32ToW16(auto_correlation_, kCorrelationLen,
                                    auto_corr, scaling);
 }
 
 bool TimeStretch::SpeechDetection(int32_t vec1_energy, int32_t vec2_energy,
                                   size_t peak_index, int scaling) const {
   // Check if the signal seems to be active speech or not (simple VAD).
   // If (vec1_energy + vec2_energy) / (2 * peak_index) <=
   // 8 * background_noise_energy, then we say that the signal contains no
   // active speech.
@@ skipping 23 matching lines @@
     int temp_scale = WebRtcSpl_NormW32(left_side);
     left_side = left_side << temp_scale;
     right_side = right_side >> (2 * scaling - temp_scale);
   } else {
     left_side = left_side << 2 * scaling;
   }
   return left_side > right_side;
 }
 
 }  // namespace webrtc