Use the AudioProcessing float interface in MediaStreamAudioProcessor.

content/renderer/media/webrtc_audio_capturer_unittest.cc

Index: content/renderer/media/webrtc_audio_capturer_unittest.cc
diff --git a/content/renderer/media/webrtc_audio_capturer_unittest.cc b/content/renderer/media/webrtc_audio_capturer_unittest.cc
index 0011f7630597f323e2cb07522df7e4e269c3f1ef..274fced6ed975d7fe015a88832af31080d51c7d2 100644
--- a/content/renderer/media/webrtc_audio_capturer_unittest.cc
+++ b/content/renderer/media/webrtc_audio_capturer_unittest.cc
@@ -130,7 +130,8 @@ class WebRtcAudioCapturerTest : public testing::Test {
     EXPECT_CALL(*sink, FormatIsSet());
     EXPECT_CALL(*sink,
                 OnDataCallback(_, _, delay_ms, expected_volume_value,
-                               need_audio_processing, key_pressed));
+                               need_audio_processing, key_pressed))
+        .Times(AtLeast(1));

[ajm, 2014/08/14 03:33:05]

The error is that this was getting called more than once on Android. I think
that's as it should be actually and the previous behavior is suboptimal. This
test on Android happens to be using 20ms frames (see above). Since
MediaStreamAudioProcessor is designed to divide the input into 10ms chunks, we
should actually be getting two 10ms callbacks on Android.

I think this wasn't happening before due to buffering in the resampler. On the
first pass, it won't provide as much output as it gets input, because it
requires a bit of delay for the filter kernel. MSAP is now just using a simple
FIFO and doesn't require the delay.

     callback->Capture(audio_bus.get(), delay_ms, volume, key_pressed);
 
     // Verify the cached values in the capturer fits what we expect.