Ensure that an RTP audio level of 127 represents digital silence.

To quote rfc6464:
  The audio level for digital silence -- for a muted audio source, for
  example -- MUST be represented as 127 (-127 dBov), regardless of the
  dynamic range of the encoded audio format.

The behavior in webrtc is correct that digital silence is represented
with 127, but it is also possible to get a value of 127 for not quite
digitally silent audio buffer (as in, not strictly 0s).

Bug: webrtc:14029
Change-Id: I7ff8698a7e4d5c0960c667fd1cc961838e269456
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/261244
Commit-Queue: Tomas Gunnarsson <tommi@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#36793}

modules/audio_processing/rms_level.cc

@@ -101,8 +101,18 @@ void RmsLevel::AnalyzeMuted(size_t length) {
 }
 
 int RmsLevel::Average() {
-  int rms = (sample_count_ == 0) ? RmsLevel::kMinLevelDb
-                                 : ComputeRms(sum_square_ / sample_count_);
+  const bool have_samples = (sample_count_ != 0);
+  int rms = have_samples ? ComputeRms(sum_square_ / sample_count_)
+                         : RmsLevel::kMinLevelDb;
+
+  // To ensure that kMinLevelDb represents digital silence (muted audio
+  // sources) we'll check here if the sum_square is actually 0. If it's not
+  // we'll bump up the return value to `kInaudibleButNotMuted`.
+  // https://datatracker.ietf.org/doc/html/rfc6464
+  if (have_samples && rms == RmsLevel::kMinLevelDb && sum_square_ != 0.0f) {
+    rms = kInaudibleButNotMuted;
+  }
+
   Reset();
   return rms;
 }