VVoice - Voice to Terminal

Speak commands in your native language. VVoice transcribes, translates, and sends them straight to your VS Code terminal.

Toggle listening: Cmd+Shift+V (macOS) / Ctrl+Shift+V (Windows/Linux)

How It Works

Microphone -> Soniox STT -> Stop Word Detection -> LLM Correction -> Terminal

Press Cmd+Shift+V (macOS) or Ctrl+Shift+V (Windows/Linux) to start listening
Speak your command in any supported language (e.g. Vietnamese, English, Chinese, Japanese...)
Say your stop word (default: "thank you") to trigger
VVoice transcribes, optionally corrects/translates via LLM, and sends the command to your active terminal
Press Cmd+Shift+V / Ctrl+Shift+V again to stop listening

Works great with AI coding assistants like Claude Code, Copilot CLI, or any terminal app.

Real-time speech-to-text via Soniox STT (streaming WebSocket)
Multilingual — speak in your native language, get commands in your target language
LLM correction — optional translation and command correction via xAI Grok
Stop word detection — say "thank you" (configurable) to send the command
Continuous listening — send multiple commands without re-toggling
Terminal targeting — choose which terminal receives commands
Status bar integration — see mic state and live transcript in the status bar
Cross-platform — works on macOS, Windows, and Linux

SoX — audio capture tool

# macOS
brew install sox

# Ubuntu/Debian
sudo apt install sox

# Windows
choco install sox

Command	Keybinding	Description
`VVoice: Toggle Voice Listening`	`Cmd+Shift+V` / `Ctrl+Shift+V`	Start/stop listening
`VVoice: Run Setup Wizard`	—	Re-run the initial setup
`VVoice: Set Soniox API Key`	—	Update your Soniox key
`VVoice: Set xAI API Key`	—	Update your xAI key
`VVoice: Select Target Terminal`	—	Pick which terminal receives commands
`VVoice: Reset All API Keys`	—	Clear all saved credentials

Setting	Default	Description
`vvoice.nativeLanguage`	`vi`	Your native language code
`vvoice.targetLanguage`	`en`	Target language for output
`vvoice.stopWord`	`thank you`	Phrase that triggers sending the command
`vvoice.autoSend`	`true`	Auto-send command on stop word
`vvoice.skipLlmCorrection`	`false`	Skip LLM and send raw transcript
`vvoice.llmModel`	`grok-4-fast-non-reasoning`	xAI model for correction
`vvoice.sonioxModel`	`stt-rt-v4`	Soniox STT model
`vvoice.terms`	`[]`	Custom recognition terms for better accuracy
`vvoice.translationTerms`	`[]`	Custom translation pairs `[{source, target}]`

You're in VS Code with Claude Code open in the terminal.

Press Cmd+Shift+V (or Ctrl+Shift+V on Windows/Linux) — status bar shows "Listening..."
Say (in Vietnamese): "chay pytest cho folder tests, thank you"
VVoice transcribes and corrects to: pytest tests/
Command is sent to the terminal and executed

Any language supported by Soniox STT, including: Vietnamese, English, Chinese, Japanese, Korean, French, German, Spanish, and many more.

Audio is streamed directly to Soniox for transcription — no audio is stored locally or by the extension
API keys are stored securely in VS Code's SecretStorage (OS keychain)
LLM correction (if enabled) sends only the text transcript to xAI, not audio

MIT