Claude Voice

Talk to Claude instead of typing. Press Ctrl+Alt+V, speak your prompt, and it appears in Claude Code's chat input — ready to send.

No API key. No cloud. No subscription. Runs 100% offline on your machine.

Why Claude Voice?

• No API key required — just install and speak
• Your voice never leaves your computer — speech is processed locally on your machine
• Works offline — no internet connection needed after the initial model download
• Free to use — no per-request costs, no subscriptions

Setup

1. Install Python dependencies

pip install openai-whisper sounddevice pynput pyperclip numpy

The first use downloads a ~140 MB speech recognition model. After that, everything runs fully offline with no internet connection required.

2. Reload VS Code

Ctrl+Shift+P → Developer: Reload Window

That's it. Look for the 🎤 Voice button in your status bar.

How to use

1. Press Ctrl+Alt+V (Windows/Linux) or Cmd+Alt+V (Mac) — or click 🎤 Voice in the status bar
2. Speak your prompt naturally
3. Pause for ~1 second when done — your words appear in Claude Code's chat input
4. Press Enter to send

Press Ctrl+Alt+V again at any time to cancel.

Requirements

• Claude Code extension installed
• Python 3.9 or higher
• A microphone

Settings

Troubleshooting

"Missing Python dependencies" — Run pip install openai-whisper sounddevice pynput pyperclip numpy in your terminal.

Nothing appears after speaking — Make sure the Claude Code panel is open. The extension automatically focuses the Claude Code input before pasting.

Text goes to the wrong app — The extension checks that VS Code is the active window before pasting. If you switch away while it's transcribing, the paste is safely discarded.

macOS — pynput requires Accessibility permissions: System Settings → Privacy & Security → Accessibility → enable VS Code.

Linux — Requires xdotool: sudo apt install xdotool

Slow first transcription — Whisper loads on first use (~5s). Subsequent transcriptions are faster.