Code by Voice: Voice Mode in Claude Code

TL;DR Enable /voice, hold the spacebar and speak. Claude Code transcribes in streaming — you see words appearing as you talk. Release and the prompt is ready. Mix voice and typing in the same message.

There's a clear trend in the AI agent world: voice interaction. Typing long, descriptive prompts in a terminal is slow. Dictating them is more natural, faster, and keeps your hands free to browse code or documentation in another window.

Claude Code has incorporated this with streaming transcription — you don't wait until you finish speaking to see text. Words appear in real time as you dictate, optimized for development vocabulary: regex, OAuth, JSON, localhost are transcribed correctly. It even picks up your project name and current git branch as recognition hints.

In my experience it still works a bit rough around the edges — sometimes the "hold to record" detection lags, or the transcription doesn't nail very specific terms. But it's good to have as an option, and it will improve.

Result:

> /voice
Voice mode enabled. Hold Space to record. Dictation language: en

> [hold Space, speak: "refactor the auth middleware
   to use the new token validation helper"]

> refactor the auth middleware to use the new
  token validation helper▮

Setup

1. Enable Voice Mode

/voice

Persists across sessions. Run /voice again to disable.

2. Speak

Hold Space and talk. You'll see keep holding… briefly, then a live waveform. Release to finalize.

Text inserts at the cursor position — you can dictate at any point in the prompt and combine with typed text.

3. Change the dictation language

Dictation uses your language setting. If not defined, defaults to English:

{
  "language": "es"
}

4. Rebind the push-to-talk key (recommended)

Space has a warmup because it needs to distinguish between holding and typing. A modifier combination starts instantly:

// ~/.claude/keybindings.json
{
  "bindings": [
    {
      "context": "Chat",
      "bindings": {
        "meta+k": "voice:pushToTalk",
        "space": null
      }
    }
  ]
}

Reference

Official docs: Voice dictation

Requirements

• Claude Code v2.1.69+
• claude.ai account (doesn't work with API key, Bedrock, Vertex, or Foundry)
• Local microphone access (doesn't work over SSH or remote environments)
• macOS/Linux/Windows (Linux may need SoX or ALSA utils)