FEAT(client): Add interaural delay

There is a small time delay (interaural time delay or ITD) between your
ears depending on the sound source position on the horizontal plane and
the distance between your ears. This commit will add this delay by using
the extra sound data in the audio buffer.

For me at least, implementing this makes it much easier to identify
where a sound is coming from when positional audio is enabled.

Implements mumble-voip#2324

src/mumble/Audio.h

@@ -17,6 +17,9 @@
 
 #define SAMPLE_RATE 48000
 
+// interaural delay (in samples) for a sound coming directly from the side of the head
+#define INTERAURAL_DELAY 0.0006 / (1 / static_cast< float >(SAMPLE_RATE))
+
 typedef QPair< QString, QVariant > audioDevice;
 
 class LoopUser : public ClientUser {

src/mumble/AudioOutput.cpp

@@ -598,6 +598,7 @@ bool AudioOutput::mix(void *outbuff, unsigned int frameCount) {
 					const float old   = (aop->pfVolume[s] >= 0.0f) ? aop->pfVolume[s] : str;
 					const float inc   = (str - old) / static_cast< float >(frameCount);
 					aop->pfVolume[s]  = str;
+					const unsigned int offset = INTERAURAL_DELAY * (1.0 + dot) / 2.0;
 					/*
 										qWarning("%d: Pos %f %f %f : Dot %f Len %f Str %f", s, speaker[s*3+0],
 					   speaker[s*3+1], speaker[s*3+2], dot, len, str);
@@ -607,10 +608,10 @@ bool AudioOutput::mix(void *outbuff, unsigned int frameCount) {
 							if (speech && speech->bStereo) {
 								// Mix stereo user's stream into mono
 								// frame: for a stereo stream, the [LR] pair inside ...[LR]LRLRLR.... is a frame
-								o[i * nchan] += (pfBuffer[2 * i] / 2.0 + pfBuffer[2 * i + 1] / 2.0)
+								o[i * nchan] += (pfBuffer[2 * i + offset] / 2.0 + pfBuffer[2 * i + offset + 1] / 2.0)
 												* (old + inc * static_cast< float >(i));
 							} else {
-								o[i * nchan] += pfBuffer[i] * (old + inc * static_cast< float >(i));
+								o[i * nchan] += pfBuffer[i + offset] * (old + inc * static_cast< float >(i));
 							}
 						}
 					}