Adding typing detection to the APM in chrome.

content/renderer/media/media_stream_audio_processor.cc

Index: content/renderer/media/media_stream_audio_processor.cc
diff --git a/content/renderer/media/media_stream_audio_processor.cc b/content/renderer/media/media_stream_audio_processor.cc
index 4422b506233fc37398e9ed1aa72bf6c341694368..0ffd3d19dc209bec1fc3d0f717b0f6096990236e 100644
--- a/content/renderer/media/media_stream_audio_processor.cc
+++ b/content/renderer/media/media_stream_audio_processor.cc
@@ -15,6 +15,7 @@
 #include "media/base/channel_layout.h"
 #include "third_party/WebKit/public/platform/WebMediaConstraints.h"
 #include "third_party/libjingle/source/talk/app/webrtc/mediaconstraintsinterface.h"
+#include "third_party/webrtc/modules/audio_processing/typing_detection.h"
 
 namespace content {
 
@@ -143,7 +144,8 @@ MediaStreamAudioProcessor::MediaStreamAudioProcessor(
     const blink::WebMediaConstraints& constraints,
     int effects)
     : render_delay_ms_(0),
-      audio_mirroring_(false) {
+      audio_mirroring_(false),
+      typing_detected_(false) {
   capture_thread_checker_.DetachFromThread();
   render_thread_checker_.DetachFromThread();
   InitializeAudioProcessingModule(constraints, effects);
@@ -264,7 +266,8 @@ void MediaStreamAudioProcessor::InitializeAudioProcessingModule(
 
   // Return immediately if no audio processing component is enabled.
   if (!enable_aec && !enable_experimental_aec && !enable_ns &&
-      !enable_high_pass_filter && !enable_typing_detection && !enable_agc) {
+      !enable_high_pass_filter && !enable_typing_detection && !enable_agc &&
+      !audio_mirroring_) {
     return;
   }
 
@@ -284,8 +287,15 @@ void MediaStreamAudioProcessor::InitializeAudioProcessingModule(
   if (enable_high_pass_filter)
     EnableHighPassFilter(audio_processing_.get());
 
-  if (enable_typing_detection)
+  if (enable_typing_detection) {
     EnableTypingDetection(audio_processing_.get());

[Henrik Grunell, 2014/02/19 06:54:42]

The semantics/naming is a bit unclear. EnableTypingDetection() doesn't really
_enable_ typing detection, it sets up |audio_processing_| so that typing
detection can get the information it needs. Which is enable VAD. What do you
think about simply renaming that function to EnableVoiceActivityDetection()?

Alternatively, have EnableTypingDetection() take both |audio_processing_| and
|typing_detector_| as parameters and do typing_detector_->SetParameters() there.

[no longer working on chromium, 2014/02/20 12:44:13]

The naming of EnableTypingDetection is corresponding to kTypingNoiseDetection
constraint, and it does not need to know the details of what APM is doing.

So I think we should keep the existing name, and move typing_detector_ inside
EnableTypingDetection.

+    // TODO(xians): Remove this |typing_detector_| after the typing suppression
+    // is enabled by default.
+    typing_detector_.reset(new webrtc::TypingDetection());
+
+    // Configure the update period to 100ms (10 * 10ms) in the typing detector.

[Henrik Grunell, 2014/02/19 06:54:42]

Nit: space between number and "ms".

[no longer working on chromium, 2014/02/20 12:44:13]

Discussed offline, in Chrome, it is very common to take the number and ms
together.

+    typing_detector_->SetParameters(0, 0, 0, 0, 0, 10);
+  }
 
   if (enable_agc)
     EnableAutomaticGainControl(audio_processing_.get());
@@ -398,6 +408,14 @@ int MediaStreamAudioProcessor::ProcessData(webrtc::AudioFrame* audio_frame,
     // TODO(xians): Swap the stereo channels after switching to media::AudioBus.
   }
 
+  if (typing_detector_ &&
+      audio_frame->vad_activity_ != webrtc::AudioFrame::kVadUnknown) {
+    bool vad_active =
+        (audio_frame->vad_activity_ == webrtc::AudioFrame::kVadActive);
+    // TODO(xians): Pass this |typing_detected_| to peer connection.
+    typing_detected_ = typing_detector_->Process(key_pressed, vad_active);
+  }
+
   // Return 0 if the volume has not been changed, otherwise return the new
   // volume.
   return (agc->stream_analog_level() == volume) ?