Added configurable offsets to the per-packet overhead in ANA.

webrtc/modules/audio_coding/audio_network_adaptor/include/audio_network_adaptor_config.h

Index: webrtc/modules/audio_coding/audio_network_adaptor/include/audio_network_adaptor_config.h
diff --git a/webrtc/modules/audio_coding/audio_network_adaptor/include/audio_network_adaptor_config.h b/webrtc/modules/audio_coding/audio_network_adaptor/include/audio_network_adaptor_config.h
index 5fc68e65222897e871d06f826ec604b95d387d9a..89878b7e457dba46d920d5c5c527f9ec43bf8446 100644
--- a/webrtc/modules/audio_coding/audio_network_adaptor/include/audio_network_adaptor_config.h
+++ b/webrtc/modules/audio_coding/audio_network_adaptor/include/audio_network_adaptor_config.h
@@ -31,6 +31,11 @@ struct AudioEncoderRuntimeConfig {
   // better use of the bandwidth. |num_channels| sets the number of channels
   // to encode.
   rtc::Optional<size_t> num_channels;
+
+  // This is true if the last frame length change was an increase, and otherwise
+  // false. Note that the default value of this variable may need to be updated

[ossu, 2017/09/14 12:16:14]

"Note that the default value of this variable may need to be updated if the
default frame length is increased."

Why?

[ivoc, 2017/09/14 13:03:39]

I guess this is a bit arbitrary and probably doesn't make a big difference, so I
decided to remove this comment. 

My thinking was that we're currently initializing the frame length to the
smallest supported value (20 ms), so it would be weird if we initialize this
variable to say that the last operation was an increase. But in the future, if
the initial frame length would be the maximum value (I guess that's pretty
unlikely anyway), then this current default wouldn't make sense anymore.

+  // if the default frame length is increased.
+  bool last_fl_change_increase = false;
 };
 
 }  // namespace webrtc