Stream speech audio to server as it gets recorded, instead of waiting until end of recording.

content/browser/speech/speech_recognizer.h

Index: content/browser/speech/speech_recognizer.h
diff --git a/content/browser/speech/speech_recognizer.h b/content/browser/speech/speech_recognizer.h
index f8df0c20312bb6ea9d5544766b3313752f99defe..51abb3bc8d6bf4a220811a397dcdc82385d77e8f 100644
--- a/content/browser/speech/speech_recognizer.h
+++ b/content/browser/speech/speech_recognizer.h
@@ -114,7 +114,6 @@ class SpeechRecognizer
   static const int kEndpointerEstimationTimeMs;
 
  private:
-  void ReleaseAudioBuffers();
   void InformErrorAndCancelRecognition(ErrorCode error);
   void SendRecordedAudioToServer();
 
@@ -137,6 +136,7 @@ class SpeechRecognizer
   Endpointer endpointer_;
   int num_samples_recorded_;
   float audio_level_;
+  std::string previous_audio_chunk_;
 
   DISALLOW_COPY_AND_ASSIGN(SpeechRecognizer);
 };