Use the route#
Prepare a recognized task context and confirm voice readiness. Hold Fn, speak a concise instruction and release. Wait for transcription, read the resulting text and make any corrections before delivery.
Start with a normal-volume, non-sensitive sentence. Once it works, evaluate softer speech separately. This avoids confusing first-use permission problems with whisper performance.
Check the result#
The development record includes real-user acceptance of hold/release recording, transcription, draft insertion and deliberate sending. That acceptance does not automatically cover cancellation, selected-text replacement, recording during a task switch or every alternative shortcut setup.
If the trigger does not start capture, use the shortcut troubleshooting page. If recording starts but text is missing, check input and model readiness. If text appears correctly, confirm the conversation before sending. Keep those stages separate so a delivery problem does not lead you to change a working microphone configuration.