Voice Mode
Code with Your Voice
As of 2026-10-20, the official voice dictation language list has exactly 20 languages, and the language code for German is de. Claude Code voice dictation enables voice-driven AI coding collaboration, supporting 20 languages for a development experience without touching the keyboard: your speech is transcribed live into the prompt input, and you can mix dictation and typing in the same message.
What Is Claude Code Voice Mode (Voice Dictation)
Voice Mode is voice dictation in the Claude Code CLI: your speech is transcribed live into the prompt input, so you can mix speaking and typing.
Two ways to record: hold Space while you speak and release it to stop, or tap once to start and tap again to send. Transcription is tuned for coding vocabulary, and your current project name and git branch name are added as recognition hints automatically.
Setup Guide
Voice dictation needs a claude.ai sign-in; it is unavailable when Claude Code is configured with an Anthropic API key or another API sign-in. The microphone has to be on your own machine, so dictation does not work in cloud sessions or SSH sessions. There are two ways to enable it: run /voice in a session, or set enabled and mode in the voice object of your user settings file.
Step 1: Install Claude Code
Install the official Claude Code CLI. As of 2026-10-20, the top entry of the official CHANGELOG is version 2.1.303.
npm install -g @anthropic-ai/claude-code@latestStep 2: Run /voice and allow microphone access
Type /voice in a Claude Code session to enable voice dictation. The first enable runs a microphone check, and on macOS your terminal gets the system microphone permission prompt.
claude config set --global voice_mode trueStep 3: Start Voice Mode
Hold mode is the default: hold Space while you speak and release it to stop; run /voice tap for tap mode instead. The dictation key is bound to voice:pushToTalk and defaults to Space, and you can rebind it to a combination such as meta+k. Caps Lock cannot be bound, because terminals do not deliver it as a key event.
Step 4: Set the dictation language
The dictation language comes from the language setting: put a language name or a BCP 47 language code in /config or in your settings file; when the setting is empty, dictation defaults to English. If the value is not on the supported list, /voice warns you and falls back to English for dictation, while Claude's text responses are not affected.
Languages supported for voice dictation
Dictation supports 20 languages, covering major developer communities worldwide. Set them through the language setting with a BCP 47 language code or a language name. As of 2026-10-20, the official voice dictation language list names all 20 languages: Czech, Danish, Dutch, English, French, German, Greek, Hindi, Indonesian, Italian, Japanese, Korean, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Turkish, Ukrainian.
Typical Voice Dictation Workflows
Dictation suits describing requirements and chasing down bugs. Adapt the sample phrases to your project; for exact paths and identifiers, typing still works better.
Code Review
Describe review requirements by voice, AI automatically analyzes code quality, security issues, and optimization suggestions
Architecture Discussion
Verbally describe system architecture needs, AI generates design proposals and implementation code
Debug Assistance
Describe bug symptoms by voice, AI identifies root cause and provides fix solutions
Documentation Writing
Dictate documentation content, AI auto-generates formatted technical docs and comments
Best Practices
When transcription is inaccurate, the key does not react, or a recording stops early, the official documentation lists the usual causes.
The microphone has to be local
Voice dictation needs a local microphone, so it does not work in cloud sessions or SSH sessions. The Claude Code VS Code extension supports dictation as well, but not in remote sessions such as SSH, Dev Containers, or Codespaces.
Recording backends differ per platform
Recording uses a built-in native module on macOS, Linux, and Windows. On Linux, when the native module cannot load, Claude Code falls back to arecord from ALSA utils or rec from SoX, and /voice prints an install command if neither is available.
WSL needs the PulseAudio backend
WSLg routes audio through PulseAudio instead of an ALSA device, so SoX needs its PulseAudio backend installed explicitly: the packages sox and libsox-fmt-pulse. Installing sox alone pulls in the ALSA backend, and WSL has no /dev/snd device, so recording does not work there.
Organization policy can disable dictation
An administrator policy for your organization can turn voice dictation off. If /voice reports that a policy blocks it, contact your organization administrator to confirm whether voice dictation is available for your organization.
Cancelling and automatic stop
Press Esc or Ctrl+C to cancel a dictation: the transcript is discarded and the prompt returns to what it held before. Recording also stops automatically after 15 seconds of silence or 2 minutes in total.
Recognition quality and auto-submit
Transcription is tuned for coding vocabulary, and your current project name and git branch name are added as recognition hints automatically. A transcript is only submitted automatically at 3 words or more; languages written without spaces, such as Japanese, Chinese, and Thai, count single words. The official CHANGELOG records the fix for that auto-submit behaviour.
Experience Voice Coding
Voice dictation needs a claude.ai sign-in and is not available with a third-party API key; transcription does not consume Claude messages or tokens, models are billed per token, and per-model rates are listed on /models.
Updated 2026-10-20