Troubleshooting the microphone module
The module fails to start when transcription is enabled
Cause. EnableTranscription loads the Whisper model at session start. A missing or
unreadable model file fails module startup outright: Whisper model not found: {Path} in the
log, not a silent fallback to voice-activity-only.
Fix. Confirm ModelPath points at a real, readable GGML model file. The default,
models/ggml-base.bin, ships bundled inside the module package; a manually relocated or replaced
model needs the setting updated to match.
Voice activity triggers without anyone speaking
Cause. With CaptureMode set to MicPlusDesktopSeparated, the voice-activity threshold is
compared against whichever is louder at that instant, the microphone or the desktop’s own audio
output. A notification sound, a call ringtone, or another participant’s voice coming through
speakers can start and end an “utterance” with the microphone completely silent.
Fix. This is expected behavior for that capture mode, not a fault. Use plain MicOnly if
only the physical microphone should ever trigger detection.
No audio or transcript is ever attached to an utterance
Cause. Both EnableAudioInComms and EnableTranscription default to off. With both off, no
audio capture stream is opened at all - only the peak meter - so there is nothing to attach even
though start/end events are still published normally.
Fix. Enable whichever of the two you need. See Overview for the privacy consideration before doing so.
The microphone works, but the module logs “exclusive use” and stops capturing
Cause. Another application opened the device in WASAPI exclusive mode, which blocks every other consumer, including this module. The module detects this specific failure and falls back to peak-meter-only monitoring rather than failing outright; voice-activity start/end events keep publishing, but no audio or transcript ever will until the other application releases the device.
Fix. Close or reconfigure whatever application holds the device exclusively, or point
DeviceId at a different capture device.
Transcription silently produces no text
Cause. A transcription failure (a malformed WAV conversion, an empty or corrupted audio buffer) is caught and logged only at debug level; the module publishes no transcript event for that utterance, with no user-facing error.
Fix. Enable DebugLogs and check for “Transcription failed” in the log around the affected
utterance.