Keep capture device running for the whole voice session

Mute now zeroes captured gain and disables the sender tracks instead of
stopping the capture device, so unmuting no longer cold-starts the AEC
(no hiss) and Bluetooth devices no longer drop/restart as they switch
between mono and stereo.

Capture is gated on voice being enabled rather than on mute: it starts
when voice is enabled and runs across calls and mute/unmute, and is
released when voice is disabled (setVoiceEnabled). Playout stays gated on
there being a connection to render. As a result the OS "mic in use"
indicator is on for the length of the session and only clears when voice
is disabled.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

1 files changed, 8 insertions, 0 deletions

diff --git a/indra/newview/llvoicewebrtc.cpp b/indra/newview/llvoicewebrtc.cpp
index ecf963039f..126d22924b 100644
--- a/indra/newview/llvoicewebrtc.cpp
+++ b/indra/newview/llvoicewebrtc.cpp
@@ -1711,6 +1711,14 @@ void LLWebRTCVoiceClient::setVoiceEnabled(bool enabled)
         mVoiceEnabled = enabled;
         LLVoiceClientStatusObserver::EStatusType status;
 
+        // Gate the audio devices on voice being enabled: the capture mic and
+        // playout speaker only run while voice is on, and the mic isn't held
+        // open when voice is off.
+        if (mWebRTCDeviceInterface)
+        {
+            mWebRTCDeviceInterface->setVoiceEnabled(enabled);
+        }
+
         if (enabled)
         {
             LL_DEBUGS("Voice") << "enabling" << LL_ENDL;