Add detection for repeated audio in capturing.

content/renderer/media/audio_repetition_detector.h

Index: content/renderer/media/audio_repetition_detector.h
diff --git a/content/renderer/media/audio_repetition_detector.h b/content/renderer/media/audio_repetition_detector.h
new file mode 100644
index 0000000000000000000000000000000000000000..58f381cd2a46fc86e6ed3a191f310849eb07ea13
--- /dev/null
+++ b/content/renderer/media/audio_repetition_detector.h
@@ -0,0 +1,141 @@
+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef CONTENT_RENDERER_MEDIA_AUDIO_REPETITION_DETECTOR_H_
+#define CONTENT_RENDERER_MEDIA_AUDIO_REPETITION_DETECTOR_H_
+
+#include <vector>
+
+#include "base/memory/scoped_vector.h"
+#include "base/threading/thread_checker.h"
+#include "content/common/content_export.h"
+
+namespace content {
+
+
+// AudioRepetitionDetector detects bit-exact audio repetitions of registered
+// patterns. A repetition pattern includes a look back time and a minimum
+// duration. The detector buffers the audio signal and checks equality of each
+// input sample against the samples at the look back positions of all registered
+// patterns, and counts the duration of any consecutive equality.
+class CONTENT_EXPORT AudioRepetitionDetector {
+ public:
+  AudioRepetitionDetector();
+  virtual ~AudioRepetitionDetector();
+
+  // Pattern is used for registering repetition patterns.
+  // A pattern includes an |id| and two time-related descriptors |look_back_ms|
+  // and |min_length_ms|, which are counted in milliseconds.
+  struct Pattern {

[ajm, 2015/09/30 00:23:43]

I didn't look closely at this, so feel free to ignore, but this looks like it
would be natural to omit id here and store a collection of Patterns in a map
rather than a vector below.

[minyue, 2015/09/30 19:52:32]

This is a good point. Yes, map is natural way to avoid id duplication, but
Pattern is a simpler, and thus I can write
const AudioRepetitionDetector::Pattern kRepetitionPatterns[] = {
  {1, 10, 10},
  {2, 20, 10}, ...

C++11's std:map initializer may make it as simple, but it is a C++11 feature. Do
you have some suggestions?

[ajm, 2015/10/01 05:13:49]

Looked more into how you're using this, and you're not looking up a Pattern by
its ID anywhere, so I don't think a map would be useful to you actually. What
you have seems fine.

[minyue, 2015/10/01 21:25:08]

Acknowledged.

+    int id;
+    int look_back_ms;
+    int min_length_ms;
+  };
+
+  // Detect repetition in |data|. A UMA report is generated upon finding
+  // a repetition. |sample_rate| is measured in Hz.
+  void Detect(const float* data, size_t num_frames, size_t num_channels,
+              int sample_rate);
+
+ private:
+  friend class AudioRepetitionDetectorForTest;  // For testing.
+
+  // A state is used by the detector to keep track of a consecutive repetition,
+  // whether the samples in a repetition are all zeros, and whether a repetition
+  // has been reported.
+  class State {
+   public:
+    State(const Pattern& pattern);
+
+    bool reported() const { return reported_; }
+    void set_reported(bool reported) { reported_ = reported; }
+
+    // Increase |count_frames_| by 1, and |zero| indidates whether the added
+    // audio frame is zero.
+    void Increment(bool zero);
+
+    // Check whether their is a valid repetition report. |sample_rate| is
+    // measured in Hz.
+    bool HasValidReport(int sample_rate) const;
+
+    void Reset();
+
+    int id() const { return pattern_.id; }
+
+    int look_back_ms() const { return pattern_.look_back_ms; }
+
+   private:
+    // The repetition pattern this state keeps track of.
+    const Pattern pattern_;
+
+    // counter of frames in a consecutive repetition.
+    size_t count_frames_;
+
+    // whether a repetition contains only zeros.
+    bool all_zero_;
+
+    // whether a repetition has been reported. This is to make sure that a
+    // repetition that contains multiple cycles will be reported at as early as
+    // the first cycle and only once.
+    bool reported_;
+  };
+
+  void RegisterRepetitionPatterns(const Pattern* patterns,
+                                  size_t num_patterns);
+
+  void ClearRepetitionPatterns();

[minyue, 2015/09/30 19:52:32]

I found this useless and therefore removed it.

+
+  // Reset |audio_buffer_| when number of channels or sample rate (Hz) changes.
+  void Reset(size_t num_channels, int sample_rate);
+
+  // Add frames (interleaved if stereo) to |audio_buffer_|.
+  void AddFramesToBuffer(const float* data, size_t num_frames);
+
+  // Determine if an audio frame (samples interleaved if stereo) is identical to
+  // |audio_buffer_| at a look back position.
+  bool Equal(const float* frame, int look_back_samples) const;
+
+  // Determine if an audio frame (samples interleaved if stereo) is zero.
+  bool IsZero(const float* frame, size_t num_channels) const;
+
+  // Action when found repetition.
+  virtual void ReportRepetition(int id);
+
+  // Used to DCHECK that we are called on the correct thread.
+  base::ThreadChecker thread_checker_;
+
+  ScopedVector<State> states_;
+
+  // The Ids of all registered patterns. This defines the range of values for
+  // the UMA report.
+  std::vector<int> ids_;
+
+  // Ring buffer to store input audio.
+  std::vector<float> audio_buffer_;
+
+  // Maximum look back time of all registered patterns. This defines the size of
+  // |audio_buffer_|
+  int max_look_back_ms_;
+
+  // Number of audio channels in buffer.
+  size_t num_channels_;
+
+  // Sample rate in Hz.
+  int sample_rate_;
+
+  // Number of frames in |audio_buffer|.
+  size_t buffer_size_frames_;
+
+  // The index of the last frame in |audio_buffer|.
+  size_t buffer_end_index_;
+
+  // The maximum frames |audio_buffer_| can take in each time.
+  size_t max_frames_;
+
+  DISALLOW_COPY_AND_ASSIGN(AudioRepetitionDetector);
+};
+
+}  // namespace content
+
+#endif  // CONTENT_RENDERER_MEDIA_AUDIO_REPETITION_DETECTOR_H_