Voice Input

Use voice input to dictate prompts to Claude Code with hold-to-talk or focus-mode recording.

Claude Code supports voice input via a hold-to-talk interface powered by Anthropic’s speech-to-text backend. Speak your prompt, release the key, and the transcript is submitted to Claude automatically.

Voice mode requires a Claude.ai account. Run /login to sign in before enabling it.

Enabling voice input

Run the /voice command to toggle voice mode on or off:

/voice

Claude Code checks the following before enabling voice:

1. Authentication — A Claude.ai account is required. If you are not logged in, you will be prompted to run /login.
2. Microphone access — The system must be able to access a recording device. On macOS and Windows, an OS permission dialog may appear.
3. Audio recording tool — Claude Code uses the native macOS audio capture module or SoX on Linux and Windows. If SoX is not installed, Claude Code shows the install command.

When voice is successfully enabled, you will see:

Voice mode enabled. Hold Space to record.

Running /voice again disables it:

Voice mode disabled.

Recording audio

Voice input uses hold-to-talk: hold the configured key while speaking, then release to submit.

• Default keybinding — Space in the Chat input context.
• Hold the key to start recording. Release to stop and send the transcript.
• A waveform visualizer shows audio levels while recording.
• Interim transcripts appear in real time as you speak.
• After you release the key, the state briefly shows processing while the transcript is finalized, then the text is injected into the prompt input.

Short taps (under 2 seconds) are silently discarded to avoid accidental activations. Hold the key for at least 2 seconds when dictating.

Supported languages

Voice input uses Anthropic’s voice_stream speech-to-text endpoint backed by Deepgram. The following languages are supported:

Language	BCP-47 code
English	en
Spanish	es
French	fr
Japanese	ja
German	de
Portuguese	pt
Italian	it
Korean	ko
Hindi	hi
Indonesian	id
Russian	ru
Polish	pl
Turkish	tr
Dutch	nl
Ukrainian	uk
Greek	el
Czech	cs
Danish	da
Swedish	sv
Norwegian	no

Setting your dictation language

Configure the dictation language via /config. The setting accepts both language names (e.g. French, Français) and BCP-47 codes (e.g. fr).

When voice is enabled, Claude Code shows the resolved language code:

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

If the configured language is not supported, Claude Code falls back to English and shows a notice:

Note: "Klingon" is not a supported dictation language; using English. Change it via /config.

Focus mode

Focus mode is an advanced recording option for multi-window workflows. When enabled, recording starts automatically whenever the Claude Code terminal gains focus and stops when it loses focus.

• Each final transcript is injected immediately, keeping the connection alive for continuous dictation.
• A silence timeout of 5 seconds tears down the session if no speech is detected. The next focus cycle re-arms recording.
• Focus mode recording is driven by window focus, not keypresses.

Enable focus mode from /config.

Tips for effective voice prompting

Speak naturally in complete sentences

The speech-to-text model performs best with complete, natural sentences. Avoid spelling out words letter-by-letter or pausing mid-sentence.

State the task clearly upfront

Start with the action: “Refactor the parseConfig function to return an error instead of throwing” rather than trailing off and adding context afterwards.

Use technical terms confidently

Claude Code sends key terms from your conversation history as hints to the STT backend to improve recognition of identifiers, function names, and project-specific vocabulary.

Hold long enough

Recordings shorter than 2 seconds are silently dropped. Hold the key for the full duration of your dictation, including a brief pause at the end before releasing.

Check your microphone input device

If you see “No audio detected from microphone”, verify that the correct input device is selected in your system audio settings and that Claude Code has microphone access.

Troubleshooting

Voice mode requires a Claude.ai account

Voice mode is only available when authenticated with a Claude.ai account. Run /login and sign in, then run /voice again.

No audio recording tool found

On Linux and Windows, Claude Code requires SoX for audio recording. Install it with:

macOS

brew install sox

Ubuntu / Debian

sudo apt-get install sox

Windows

Download SoX from https://sox.sourceforge.net/ or use winget:

winget install sox

Microphone access is denied

Claude Code needs microphone permission from the OS:

• macOS — Go to System Settings → Privacy & Security → Microphone and enable access for your terminal app.
• Windows — Go to Settings → Privacy → Microphone and allow access.
• Linux — Check your system’s audio settings and PipeWire/PulseAudio permissions.

After granting access, run /voice again.

No speech detected after recording

If you held the key for more than 2 seconds but received “No speech detected”:

• Verify your microphone is not muted at the hardware or system level.
• Check that the correct input device is selected.
• Try a different recording environment to rule out background noise issues.

Voice connection failed

A connection failure usually means the WebSocket to Anthropic’s speech-to-text backend could not be established. Check your network connection and try again. If you are behind a proxy or firewall, ensure WebSocket connections to Anthropic’s API are allowed.