Push-to-talk voice dictation that transcribes speech live into the prompt input.

Claude Code Voice Dictation

Use push-to-talk voice dictation to speak prompts instead of typing them. Speech is transcribed live into the prompt input, and you can mix voice and typing in the same message.

Requires Claude Code v2.1.69 or later.

Requirements

Authentication: Claude.ai account only. Not available with API key, Bedrock, Vertex, or Foundry.
Local microphone: Does not work in remote environments (Claude Code on the web, SSH). WSL requires WSLg (WSL2 on Windows 11).
Audio recording: Built-in native module on macOS, Linux, and Windows. On Linux, falls back to arecord (ALSA) or rec (SoX) if the native module can't load.
Transcription: Audio is streamed to Anthropic's servers. Not processed locally.

Enabling Voice Dictation

/voice
Voice mode enabled. Hold Space to record. Dictation language: en (/config to change).

Persists across sessions. Run /voice again to disable, or set in settings:

{
  "voiceEnabled": true
}

First-time use triggers a microphone check. On macOS, this prompts for terminal microphone permission.

Recording

Hold Space to start recording:

Warmup: Footer shows keep holding... while detecting held key via key-repeat
Recording: Live waveform displays, speech appears dimmed in prompt
Release: Transcript finalizes and inserts at cursor position

Single Space tap still types a space normally. Mix typing and dictation freely:

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

Transcription is tuned for coding vocabulary: regex, OAuth, JSON, localhost, etc. Your project name and git branch are added as recognition hints automatically.

Push-to-Talk Key

Default is Space. Rebind in ~/.claude/keybindings.json:

{
  "bindings": [
    {
      "context": "Chat",
      "bindings": {
        "meta+k": "voice:pushToTalk",
        "space": null
      }
    }
  ]
}

Modifier combinations (e.g., meta+k) start recording on the first keypress with no warmup. Avoid bare letter keys since they type during warmup.

Dictation Language

Uses the language setting (same as Claude's response language). Defaults to English if unset.

{
  "language": "japanese"
}

Supports BCP 47 codes or language names. If the language isn't supported, dictation falls back to English with a warning.

Supported Languages

Language	Code	Language	Code
Czech	`cs`	Korean	`ko`
Danish	`da`	Norwegian	`no`
Dutch	`nl`	Polish	`pl`
English	`en`	Portuguese	`pt`
French	`fr`	Russian	`ru`
German	`de`	Spanish	`es`
Greek	`el`	Swedish	`sv`
Hindi	`hi`	Turkish	`tr`
Indonesian	`id`	Ukrainian	`uk`
Italian	`it`	Japanese	`ja`

Troubleshooting

Issue	Fix
"Voice mode requires a Claude.ai account"	Run `/login` to sign in with Claude.ai (not API key or third-party provider)
"Microphone access is denied"	Grant microphone permission in system settings, then run `/voice` again
"No audio recording tool found" (Linux)	Install SoX: `sudo apt-get install sox`
Nothing happens when holding Space	Ensure voice is enabled (`/voice`). Check terminal sends key-repeat events
Transcription garbled or wrong language	Set correct language in `/config`

Voice Dictation