How to Use Claude Code Voice Input (/voice) in Japanese¶

For / Key Points

For: Developers who want to use Claude Code's voice input in Japanese

Key Points:

Change /config → Language → Japanese to enable Japanese voice recognition
Hold spacebar (push-to-talk) to speak, release to trigger recognition
Pronounce technical terms in English for better accuracy

You speak Japanese into Claude Code's /voice mode, but the transcription either fails silently or returns gibberish. The fix takes one setting change in /config.

The Default Setting Doesn't Recognize Japanese¶

Claude Code's /voice command launched in March 2026¹. Hold the spacebar, speak, and your voice is transcribed into text and sent as a prompt.

The problem is the language setting. The default recognition language is English. When you speak Japanese, the transcription engine interprets it as English and returns unintelligible text — or in some cases, "No speech detected."

One configuration change fixes this entirely.

The Fix: Set Language to Japanese in /config¶

The setup takes three steps.

Step 1: Open /config¶

Type /config in the Claude Code prompt and press Enter.

> /config

Step 2: Select Language¶

A settings menu appears. Use the ↓ key (down arrow) to move the cursor to the Language option.

Do not press Enter here — Enter closes the config menu. Instead, press Space to enter the Language setting.

Step 3: Type Japanese and confirm¶

A text input field appears. Type Japanese and press Enter. The speech recognition target language switches to Japanese.

The setting persists

Once configured, subsequent /voice sessions retain the Japanese setting. No need to reconfigure each time.

Usage: Voice Input with Push-to-Talk¶

With the language set, try voice input.

Basic Operation¶

Type /voice to activate voice mode
Hold the spacebar
Wait 1-2 seconds until "listening..." appears on screen⁴
Speak in Japanese (e.g., "Improve the error handling in this file")
Release the spacebar to trigger recognition

The transcribed text is inserted into the prompt, and Claude begins responding.

Exiting Voice Mode¶

Type /voice again or press Ctrl+C to exit voice mode.

Manual /voice activation required each session

There is no claude --voice CLI flag. You must run the /voice command manually at the start of each session.

5 Tips to Improve Recognition Accuracy¶

Japanese voice input works, but accuracy lags behind English. These tips help close the gap.

Pronounce technical terms in English — "refactoring" is recognized more accurately than its Japanese pronunciation. Function names and file names also benefit from English pronunciation
Break instructions into short segments — One or two sentences at a time yields better recognition than long continuous speech
Wait for the listening indicator — Speaking immediately after pressing spacebar cuts off the beginning. Wait for "listening..." to appear
Use a quiet environment — Push-to-talk reduces noise impact, but a quiet setting still improves accuracy
Mix Japanese and English naturally — Say "auth.py no validate function wo fix shite" — code elements in English, instructions in Japanese. The recognizer handles this pattern well

Limitations and Future Outlook¶

Key limitations as of March 2026:

Limitation	Details
No default voice mode startup	No `claude --voice` CLI flag exists
Unreliable auto-detection	Explicit language setting via `/config` is required
Technical term accuracy	Japanese domain-specific terms are recognized less accurately than English

GitHub Issue #31724 requests a voiceLanguage setting². Over 21 upvotes and reports from users of Ukrainian, French, Turkish, German, and other languages confirm this is a widespread need.

The proposed solution uses ISO 639-1 language codes (e.g., "voiceLanguage": ["ja", "en"]) with improved auto-detection. Once implemented, language preferences could be set in configuration files without going through /config.

Alternative: VoiceMode MCP Server¶

Beyond the native /voice, the community-built VoiceMode MCP server offers an alternative approach³. It suits developers who need always-on voice mode.

Features¶

Runs Whisper (STT) and Kokoro (TTS) locally with no API keys required
Enables two-way voice conversations (you speak, Claude responds with audio)
Smart silence detection automatically stops recording when you stop speaking

Installation¶

# Install VoiceMode
curl -LsSf https://astral.sh/uv/install.sh | sh
uvx voice-mode-install

# Add as MCP server to Claude Code
claude mcp add --scope user voicemode -- uvx --refresh voice-mode

The key difference from native /voice is that VoiceMode runs as a persistent MCP server — no need to type /voice each session. The tradeoff is that running Whisper and Kokoro locally requires meaningful machine resources.

Native /voice is sufficient for most cases

If you just want to give instructions in Japanese, changing the language setting in /config is all you need. VoiceMode MCP is for developers pursuing two-way voice conversations or fully hands-free workflows.

Voice Input Is One Setting Away¶

Japanese voice input in Claude Code is enabled by a single /config change that most developers overlook. Setup takes under a minute.

Speaking "fix this test" instead of typing it — once you experience code changing from a spoken command, it becomes hard to go back. Voice input particularly excels during code reviews and design discussions, where thinking out loud is faster than typing.

As the voiceLanguage setting moves toward official implementation, multilingual voice input including Japanese will become even more accessible.