Voice dictation: allow typing while recording is active (don't stop the recording on keystrokes)

Status Open
Maintainer reply None cached
Activity 0 comments · opened Aug 14, 2026

Feature request

Allow typing in the prompt input while a voice dictation recording is active, instead of the recording being interrupted by keystrokes.

Current behavior

With voice dictation enabled (voice.mode: "tap" or "hold"), starting to type while a recording is in progress stops/tears down the recording session. Dictation and typing can only be sequenced (type, then dictate, then type), never overlapped.

Desired behavior

Keep the recording alive while the user types. The live interim transcript already streams into the prompt input — if typing didn't kill the session, the user could correct a mis-transcription the moment they see it and keep talking.

Grok's desktop voice input works exactly this way: start recording, pause speaking, type a correction or addition mid-recording, resume speaking — the recording never stops. For long-form dictation it makes a huge practical difference.

Use case

Long spoken prompts (several minutes of context-setting). Speech-to-text inevitably mishears the odd word ("hazard" → "has ord"). Today the only options are to stop the recording, fix the word, and restart dictation — or let the wrong word (sometimes with the wrong meaning) go through to the model. Being able to reach over and fix it in real time, without losing the recording, would make extended voice sessions genuinely viable.

Related

  • #74534 (configurable silence/max-duration timeouts) — same underlying goal: making dictation usable for long-form spoken input. The 15s-silence auto-stop compounds this issue: pausing to think about a correction can itself end (and in tap mode, submit) the recording.

Environment: Claude Code v2.1.232, macOS (arm64), terminal TUI, Claude.ai (Max) auth.

View original on GitHub ↗