Per-app input device selection (dictate from a handset/Bluetooth mic without changing the macOS system default)

[aguynamedryan]

Idea

Let FluidVoice capture from a specific input device (a USB dictation handset, a Bluetooth headset, etc.) without changing the macOS system default input — so the rest of the system keeps using the built-in mic while FluidVoice uses the chosen device.

Background

Independent input selection existed but was removed in c8df54a because it was built on AVAudioEngine's input node, which fails with -10851 for aggregate/Bluetooth devices (plus 2nd-recording crashes and BT-disconnect freezes).

Since then, b75c23f added the direct Core Audio capture path (DirectCoreAudioInput → AudioDeviceCreateIOProcID), which binds an arbitrary input device per-app without touching the system default and without the AVAudioEngine -10851 limitation. So the blocker that motivated the removal no longer applies to the capture path that now ships by default.

Proposal

I've opened #642, which re-enables independent input selection routed exclusively through the direct-capture path and pins the AVAudioEngine fallback to the system default so -10851 can't recur. Validated on hardware: USB handset, generic Bluetooth (Shokz), and AirPods each captured per-app with the system default left on the built-in mic and no -10851; aggregate devices fall back gracefully.

Independent mode — FluidVoice on the handset while the macOS default stays the built-in mic:

Would you be open to this? Happy to adjust the approach — keep it behind the experimental flag, rename the setting, gate it differently. Relates to #610.