[Bug] Microphone input fails completely — AudioUnitRender error -10863 on macOS Apple Silicon (built-in mic, Bluetooth)

[GaSeDevAI]

Describe the bug

The Prismical app fails to capture audio from the microphone regardless of the input device used. I tested both the built-in MacBook Pro microphone and several headsets with microphones (including Bose QuietComfort 35 Bluetooth and other Bluetooth headsets), and the same error occurs in every case. The system audio (output) is captured correctly, but the microphone always produces empty WAV files (0 bytes, 0 duration) due to an AudioUnitRender(microphone) failed: -10863 error (kAudioUnitErr_CannotDoInCurrentContext). As a result, transcription is never generated (transcriptSegmentCount: 0).

App Version

v0.1.0

Model Information

• whisper-medium (initial, ~1.5 GB) — auto-downloaded
• whisper-tiny (manual switch for testing, ~78 MB)
• Mode: local

System Information

• Operating System: macOS 26.5.2 (Darwin arm64)
• RAM: 16 GB
• GPU: Apple M2 Pro (integrated)
• CPU: Apple M2 Pro (10 cores, 10 threads, 2.4 GHz)
• Hardware: Mac14,12 (Apple Inc.)

To Reproduce

1. Launch Prismical v0.1.0 on macOS with Apple Silicon
2. Complete onboarding and download a Whisper model (e.g., whisper-medium)
3. In settings, select a microphone (built-in, Bluetooth headset, or any other input device)
4. Start a new recording/meeting from the app
5. Speak into the microphone for a few seconds
6. Stop the recording
7. Notice that no transcription is generated

Expected behavior

Microphone audio should be captured correctly and sent to the Whisper model for transcription, producing text output regardless of the selected input device (built-in microphone, Bluetooth, USB, etc.).

Logs

Attached: prismical-logs-2026-07-06.log

Key findings from the log:

1. Recurring error: Every audio capture attempt produces repeatedly:

   [audio-capture] ERROR: AudioUnitRender(microphone) failed: -10863
   

   Error -10863 corresponds to CoreAudio's kAudioUnitErr_CannotDoInCurrentContext, indicating the microphone's AudioUnit is unable to render in the current context.

2. The issue occurs with multiple devices: The error was reproduced with:

   • Built-in microphone of the MacBook Pro M2 Pro (during initial tests at ~11:50)
   • Bose QuietComfort 35 Bluetooth headset (explicitly selected at 11:55:07)
   • Other Bluetooth headsets (subsequent tests)
     In all cases the result is identical: AudioUnitRender(microphone) failed: -10863.

3. The issue persists even after resetting the preferred microphone: At 11:56:13 the preferred microphone was set back to null, but the next capture attempt (12:02:14) still shows the same error.

4. System audio OK, microphone KO: System audio is captured successfully (system-pre-resample.wav files contain valid data), but microphone files are always empty:

   mic.wav: dataSize: 0, duration: 0
   mic_processed.wav: dataSize: 0, duration: 0
   

5. No transcription produced: Every session ends with:

   Meeting capture finalized { meetingId: '...', transcriptSegmentCount: 0 }
   

6. The issue persists across app restarts: The error repeats in separate sessions (both at ~11:50 and ~20:25), after fully restarting the application.

Representative log excerpt:

[2026-07-06 11:51:22.321] [info]  [ (audio)] [audio-capture] ERROR: AudioUnitRender(microphone) failed: -10863
[2026-07-06 11:51:22.335] [info]  [ (audio)] [audio-capture] System audio capture started
[2026-07-06 11:51:22.335] [info]  [ (audio)] [audio-capture] Dual mode capture started: aec=webrtc-aec3 ...
[...]
[2026-07-06 11:52:29.191] [info]  [ (transcription)] Finalized WAV file { path: '.../mic.wav', dataSize: 0, duration: 0 }

Additional context

• The kAudioUnitErr_CannotDoInCurrentContext (-10863) CoreAudio error on macOS typically occurs when the microphone AudioUnit is unable to render — an issue I encountered with all tested input devices (built-in microphone, Bose QuietComfort 35 Bluetooth, and other Bluetooth headsets).
• The mode: 'dual' setup with AEC (WebRTC-AEC3) creates an aggregate device that combines microphone and system audio; I suspect the aggregate device configuration created by the app may be incompatible with the microphone's capabilities, regardless of the specific device.
• Given that the issue also occurs with the built-in MacBook Pro M2 Pro microphone, this could be a more general bug in the microphone input handling within dual-capture mode on macOS with Apple Silicon chips.

[GaSeDevAI]

LOG

prismical-logs-2026-07-06.log