Claude Code Voice Dictation: Talk Instead of Typing

Claude Code voice dictation lets you speak your prompts instead of typing them: run /voice, then hold Space while you talk and your words appear live in the prompt. It needs Claude Code v2.1.69 or later and a Claude.ai account — and transcription doesn't consume a single token of your plan.

I'm Evgeny, and here's a confession: my best prompts are spoken, not typed. When you describe a problem out loud, you naturally add the context a typed one-liner leaves out. For non-programmers this is a superpower — you were never going to out-type a developer anyway, but you can absolutely out-explain one.

1How to enable voice mode in Claude Code

Run /voice once. The first time, Claude Code does a microphone check — on a Mac this triggers the system permission prompt for your terminal; click Allow. From then on, hold Space to record: your speech streams into the prompt as dimmed text, finalizes when you release, and you can keep typing around it. Voice and keyboard mix freely in the same message.

/voice

2Hold mode vs tap mode — which one should I use?

Hold is the default: push-to-talk, like a walkie-talkie. It has a brief warmup while Claude Code detects the held key, and a stray space or two may flash in the input before being cleaned up automatically. Tap mode is calmer: tap Space once to start, speak, tap again — and if you said at least three words, the prompt sends itself. Recording also stops on its own after 15 seconds of silence or two minutes total. I recommend tap mode to beginners: no warmup, no held keys, fewer surprises.

/voice tap

Can I change the dictation key?

Two refinements once you're comfortable. If Space feels awkward, the dictation key is the voice:pushToTalk action in ~/.claude/keybindings.json — rebind it to a modifier combo like meta+k, which also skips the hold-mode warmup because recording starts on the first keypress. And in hold mode you can set "autoSubmit": true in the voice settings so releasing the key sends the prompt automatically whenever the transcript is at least three words long.

3Does it understand languages other than English?

Yes — twenty languages, including Spanish, French, German, Japanese, Portuguese, Russian and Ukrainian. Dictation follows the same language setting that controls Claude's response language; if it's empty, you get English. Set it in /config or in your settings file, and one nice touch: transcription is tuned for coding vocabulary, so words like regex, OAuth, JSON and localhost come out right, and your project and branch names are added as recognition hints automatically.

{
  "language": "spanish"
}

4Is it private, and what does it cost?

Two honest facts. First: your audio is streamed to Anthropic's servers for transcription — nothing is processed locally. Second: it's free in the sense that matters — transcription does not consume Claude messages or tokens and doesn't count toward the limits in /usage. Dictate hour-long monologues; your plan won't notice.

5Does voice input work in VS Code?

Yes — the Claude Code VS Code extension supports dictation with the same Claude.ai account requirement. The one exception is remote sessions (SSH, Dev Containers, Codespaces): your microphone lives on your laptop while the extension runs on the remote machine, so they can't hear each other.

The kicker: tomorrow morning, describe your first task to Claude by voice — sloppy, with "umm" and self-corrections, like a human. Then look at how much context you gave it compared to what you usually type. That gap is why voice prompts get better answers.