Adding BlockMeanCalculator for AEC.

webrtc/modules/audio_processing/utility/mean_calculator.cc

+/*
+ *  Copyright 2016 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_processing/utility/mean_calculator.h"
+
+namespace webrtc {
+
+MeanCalculator::MeanCalculator(size_t window_length)
+    : window_length_(window_length),
+      count_(0),
+      sum_(0.0),
+      compensation_(0.0),
+      mean_(0.0) {
+}
+
+void MeanCalculator::Reset() {
+  Clear();
+  mean_ = 0.0;
+}
+
+void MeanCalculator::AddSample(float sample) {
+  KahanSum(sample);
+
+  // TODO(minyue): For a bug in AEC, the mean is calculated wrong. The following
+  // line should become "++count_ == window_length_" when bit exactness test in

[peah-webrtc, 2016/03/18 09:18:34]

What do you mean by this todo? As far as I can see, the following line is what
you state.

[minyue-webrtc, 2016/03/21 15:27:51]

Sorry. I forgot to remove this when I came up with a better solution.

+  // AEC is updated.
+  if (++count_ == window_length_) {

[peah-webrtc, 2016/03/18 09:18:34]

Please move the increement outside of the if-statement. I think it increases the
readability and it is also easier when attaching a debugger.

[minyue-webrtc, 2016/03/21 15:27:51]

Done.

+    mean_ = sum_ / window_length_;
+    Clear();
+  }
+}
+
+size_t MeanCalculator::GetNumberNewSamples() const {
+  return count_;
+}
+
+float MeanCalculator::GetLatestMean() const {
+  return mean_;
+}
+
+// Flush all samples added.
+void MeanCalculator::Clear() {
+  count_ = 0;
+  sum_ = 0.0;
+  compensation_ = 0.0;
+}
+
+void MeanCalculator::KahanSum(float sample) {

[peah-webrtc, 2016/03/18 09:18:34]

I don't think we need to compute the sum using KahanSum. That method is at least
5 times as complex as a simple add and I doubt that we need it any longer since
we are no longer doing recursive windowed averages. At any rate, I don't think
it should be there if there is no clear need for it.

[minyue-webrtc, 2016/03/21 15:27:51]

KahanSum is better in accuracy in any long-window sum, regardless of recursive
or not.

But yes, it is more complex. I can remove it, if accuracy is not a big deal (and
it should not be, since inaccuracy has been in many other places.).

+  const float compensated_sample = sample - compensation_;
+  const float temp_sum = sum_ + compensated_sample;
+  compensation_ = (temp_sum - sum_) - compensated_sample;
+  sum_ = temp_sum;
+}
+
+}  // namespace webrtc