Voice Input

Voice Input

Voice input is currently available on macOS. It uses the same Meta credential as the model provider; a separate Muse login is not required.

Record and insert a transcript

1. Press Alt+V to start recording.
2. Speak while the green microphone meter is visible in Pi's status bar.
3. Press Alt+V again to stop.
4. Review the transcript inserted into Pi's editor before submitting it.

Commands

• /voice — show status and privacy behavior
• /voice-on — enable the Alt+V shortcut
• /voice-off — disable voice input

The first recording may trigger the macOS microphone permission prompt.

Audio handling

Audio is captured only between the two Alt+V presses. The helper records 16 kHz mono PCM and streams it to Meta for transcription. No local speech-recognition model is downloaded.

Undocumented ASR dependency

The default service is an internal Muse Code Meta ASR WebSocket endpoint. It is undocumented, unversioned, and may break after a server-side change even when local tests pass.

Optional overrides:

• PI_META_VOICE_ASR_ENDPOINT
• PI_META_VOICE_ASR_MODEL

Endpoint overrides must use wss://. For local transport testing only, ws://localhost and ws://127.0.0.1 are accepted.

Common failures

• could not connect to Muse's Meta ASR service — connection setup failed
• Meta ASR connection closed (<code>) — the service closed an established connection

These messages do not distinguish a moved endpoint from a network outage. Continue with Troubleshooting.