Using the NS noise estimate for the IE

webrtc/modules/audio_processing/audio_processing_impl.cc

Index: webrtc/modules/audio_processing/audio_processing_impl.cc
diff --git a/webrtc/modules/audio_processing/audio_processing_impl.cc b/webrtc/modules/audio_processing/audio_processing_impl.cc
index 0934b1e4afa9f519fa91e42dde7439f047e009b4..71d2e21cadb65eae4c61d02d82f5de7ef565f680 100644
--- a/webrtc/modules/audio_processing/audio_processing_impl.cc
+++ b/webrtc/modules/audio_processing/audio_processing_impl.cc
@@ -771,12 +771,6 @@ int AudioProcessingImpl::ProcessStreamLocked() {
     ca->SplitIntoFrequencyBands();
   }
 
-  if (constants_.intelligibility_enabled) {
-    public_submodules_->intelligibility_enhancer->AnalyzeCaptureAudio(
-        ca->split_channels_f(kBand0To8kHz), capture_nonlocked_.split_rate,
-        ca->num_channels());
-  }
-
   if (capture_nonlocked_.beamformer_enabled) {
     private_submodules_->beamformer->ProcessChunk(*ca->split_data_f(),
                                                   ca->split_data_f());
@@ -793,6 +787,11 @@ int AudioProcessingImpl::ProcessStreamLocked() {
     ca->CopyLowPassToReference();
   }
   public_submodules_->noise_suppression->ProcessCaptureAudio(ca);
+  if (constants_.intelligibility_enabled &&
+      public_submodules_->noise_suppression->is_enabled()) {

[turaj, 2016/02/09 16:40:33]

Can we have noise estimation part of NS to be auto-enabled when IE is enabled?
Otherwise one has to know about details that how sub-modules depend on one
another.

[aluebs-webrtc, 2016/02/09 19:13:35]

I thought about pulling out the noise-estimation from the NS as a stand-alone
component, but the truth is that this is a ton of work, mainly because it is old
legacy C-code and is implemented radically different on the float and fixed
versions. To give you an idea, the float version calculates the noise-estimate
on the analysis stage, while the fixed version does it in the process stage. But
I don't think it makes sense to have yet another separate noise-estimation (AEC
has its own already), since the one in the NS gives a good estimate and runs in
most cases anyway.
I agree that these dependencies between components are sub-optimal, but we
already have some, for example the TS depends on the AGC. I think it is safe to
assume that a noisy environment that would benefit from IE would also do so from
NS.
The only solution that I see, without going into a huge refactoring project of
the NS, is to enable the NS when the IE is enabled. WDYT?
If you have another idea of how we could deal with this, please let me know.

[hlundin-webrtc, 2016/02/09 19:25:03]

You could add a (D)CHECK in a strategic position to verify that the relevant
sub-components are enabled.

[aluebs-webrtc, 2016/02/09 23:09:09]

Done.

+    public_submodules_->intelligibility_enhancer->SetCaptureNoiseEstimate(
+        public_submodules_->noise_suppression->noise_estimate());
+  }
   RETURN_ON_ERR(
       public_submodules_->echo_control_mobile->ProcessCaptureAudio(ca));
   public_submodules_->voice_detection->ProcessCaptureAudio(ca);