Fix for integer overflow in NetEq.

webrtc/modules/audio_coding/neteq/expand.cc

Index: webrtc/modules/audio_coding/neteq/expand.cc
diff --git a/webrtc/modules/audio_coding/neteq/expand.cc b/webrtc/modules/audio_coding/neteq/expand.cc
index ffe4370e4e61e8766c50ce8efcf22c96bd9723bf..4c13b689a0a15ef0b35b6a08bf6f4b9f5abea077 100644
--- a/webrtc/modules/audio_coding/neteq/expand.cc
+++ b/webrtc/modules/audio_coding/neteq/expand.cc
@@ -676,7 +676,10 @@ void Expand::AnalyzeSignal(int16_t* random_vector) {
                               kUnvoicedLpcOrder + 1,
                               128);
     int16_t unvoiced_prescale;
-    if (WebRtcSpl_MaxAbsValueW16(unvoiced_vector, 128) > 4000) {
+    int16_t unvoiced_max_abs = WebRtcSpl_MaxAbsValueW16(unvoiced_vector, 128);
+    if (unvoiced_max_abs > 16383) {
+      unvoiced_prescale = 6;

[kwiberg-webrtc, 2016/12/13 13:09:46]

So, let's see... you sum 2^7 values that are all less than 2^30, and divide by
2^6, so the shifted sum ends up being less than 2^31. Sounds right.

+    } else if (unvoiced_max_abs > 4000) {
       unvoiced_prescale = 4;

[kwiberg-webrtc, 2016/12/13 13:09:46]

Sum of 2^7 values that are all less than 2^28, and divide by 2^4, giving a sum
less than 2^31. Good.

     } else {
       unvoiced_prescale = 0;

[kwiberg-webrtc, 2016/12/13 13:09:46]

Sum of 2^7 values that are all less than 2^24 (actually, less than 16000000),
giving a sum less than 2^31. Check.

So, good. But why not simply calculate unvoiced_prescale as 31 - 7 -
2*bits_in_max_element? I know we do that in other places?