How to Use Voice Dictation in Claude Code

The short answer: Claude Code has dictation built in. Type /voice, then hold Space while you talk and let go when you’re done. Your words appear in the prompt and you press Enter to send. It needs a Claude.ai sign-in and an internet connection, and it only works inside Claude Code. If you sign in with an API key, work over SSH, or want the same voice typing in every other app too, a dictation hotkey like HyperVoice does the job instead.

Option 1: Claude Code’s built-in /voice

Anthropic added voice dictation to the Claude Code CLI, and it’s good. To use it:

1. Run /voice. Claude Code checks your microphone and replies Voice mode enabled (hold). Hold space to record.
2. Hold Space and speak. After a brief warm-up the footer shows listening… and your words appear in the prompt as you talk.
3. Let go of Space. The text is finalised at your cursor. Read it, then press Enter.

A few things that make it pleasant for coding, all from Anthropic’s documentation:

• It knows developer words. Terms like regex, OAuth, JSON and localhost are recognised, and your project name and git branch are added as hints.
• You can mix voice and typing. Type half a prompt, hold Space to dictate the rest, move the cursor and dictate somewhere else.
• Tap mode sends for you. /voice tap turns Space into tap-to-start, tap-to-send. Prompts of three words or more go straight to Claude.
• It doesn’t cost usage. Transcription doesn’t count against your Claude messages or tokens.

What it needs (and where it stops)

This is where people get stuck, so it’s worth knowing before you rely on it:

• A Claude.ai account sign-in. If Claude Code is set up with an Anthropic API key, Amazon Bedrock, Google Cloud or Microsoft Foundry, /voice isn’t available, and you’ll see Unknown command: /voice.
• The internet. Your audio is streamed to Anthropic’s servers to be transcribed. It isn’t processed on your machine.
• A local microphone. It doesn’t work in SSH sessions or cloud sessions. In WSL it needs WSLg.
• Claude Code itself. It types into Claude Code’s prompt (the CLI, or the VS Code extension). Your PR description, the Slack thread and the ticket you’re about to write still need typing.

If none of that bothers you, /voice is the simplest option and you don’t need anything else.

Option 2: A dictation hotkey (HyperVoice)

HyperVoice is a dictation hotkey for Windows: press it in any app, talk, press it again, and the text lands where your cursor is. Claude Code is one of those apps.

1. Install it from the Microsoft Store. It’s signed by Microsoft, so there’s no install warning, and setup ends with a test dictation.
2. Click into Claude Code’s prompt, in Windows Terminal, VS Code’s terminal, or wherever you run it.
3. Press Ctrl+Shift+Space and explain the task the way you would to a colleague: what to change, what to leave alone, why.
4. Press Ctrl+Shift+Space again. The prompt is pasted in. Read it, then press Enter.

What’s different from /voice:

• It works however Claude Code is signed in. API key, Bedrock, Claude.ai: it doesn’t matter, because HyperVoice is just typing into the prompt.
• Speech-to-text runs on your PC (Whisper or NVIDIA Parakeet models), so your audio isn’t sent anywhere to be transcribed. The optional clean-up styles, which tidy what you said into a message or bullet points, are off by default; when you turn one on, the text (never the audio) goes to an AI provider.
• It works over SSH and in WSL. Recognition happens on Windows and only the finished text is pasted into the terminal, so it doesn’t matter what the terminal is connected to.
• It works everywhere else. The same hotkey types your commit message, the PR description, the reply in Slack and the email to the client.

It’s free at 500 words a day with a free account, no card.

Let it press Enter: “send it”

HyperVoice can submit the prompt for you, like /voice tap mode. Turn it on in Settings → Dictation → Voice commands, then end a dictation with the words “send it”. Only the very end of a dictation counts, so saying “send it” in the middle of a sentence doesn’t submit anything.

How to talk to Claude Code

Whichever tool you use, voice changes what you put in the prompt more than how fast it arrives. Typed prompts come out short because typing a paragraph is effort. Spoken ones come out complete:

• Give the context you’d normally skip. “The login page breaks on Safari, I think it’s the date parsing in the session helper, don’t touch the API routes.”
• Say what done looks like. “It’s done when the existing tests pass and there’s a test for the Safari case.”
• Type the exact bits. File paths, flags and identifiers are quicker to type than to spell out. Dictate the sentence, then type src/lib/session.ts where it belongs.

There’s more on voice in editors and coding agents in Voice Typing in VS Code, Cursor, and Claude Code, and on dictating at the prompt in general in How to Voice Type in Windows Terminal.

Which one to use

If you’re signed in to Claude Code with a Claude.ai account, work locally and only want to talk to Claude, start with /voice: it’s built in and doesn’t touch your Claude usage. If you use an API key or Bedrock, work over SSH, would rather your audio stayed on your PC, or want one hotkey for Claude Code and everything else you write, HyperVoice is built for that. Try it free from the Microsoft Store.