Speak Claude - Voice to Text for VS Code

🎤 Voice-to-text extension for VS Code and Claude Code using local WhisperX transcription. Privacy-first, runs entirely on your machine!

Want to skip the local setup? Try Speak Claw — our cloud-powered version. No Docker, no WhisperX, no local transcription service. Just install, add your API token, and start speaking. Get Speak Claw →

Quick Start (self-hosted):

1. Install this extension from VS Code Marketplace
2. Clone the repo: git clone https://github.com/SlyCreator/speak-claude.git
3. Run WhisperX: cd speak-claude && docker-compose up -d
4. Press Cmd+Shift+C anywhere in VS Code to dictate!

Features

• 🎤 Voice Recording: Press Cmd+Shift+C (Mac) or Ctrl+Shift+C (Windows/Linux) to start/stop recording
• 🤖 WhisperX Integration: Uses your local WhisperX service for high-quality transcription
• 💬 Works Everywhere: Inserts text into any VS Code editor, terminal, or input field (including Claude Code chat)
• ⚙️ Configurable: Set WhisperX URL, language, and diarization options

Prerequisites

⚠️ IMPORTANT: This extension requires the WhisperX service to be running locally. You must clone the repository and set up WhisperX before using this extension.

1. Clone the Repository

First, clone the Speak Claude repository to get the WhisperX service:

git clone https://github.com/SlyCreator/speak-claude.git
cd speak-claude

2. Set Up WhisperX Service

Option A: Docker (Recommended)

docker-compose up -d

The service will start at http://localhost:48001

Option B: Manual Python Setup

cd whisperx-service
pip install -r requirements.txt
pip install whisperx
uvicorn main:app --host 0.0.0.0 --port 48001

3. Install SoX (Sound eXchange)

The extension uses SoX for audio recording. Install it:

macOS:

brew install sox

Ubuntu/Debian:

sudo apt-get install sox libsox-fmt-all

Windows: Download from SoX SourceForge

Installation

Install from VS Code Marketplace (Recommended)

1. Open VS Code
2. Go to Extensions (Cmd+Shift+X or Ctrl+Shift+X)
3. Search for "Speak Claude"
4. Click Install

Or install via command line:

code --install-extension slycreator.speak-claude

After Installation

Make sure you have:

1. ✅ Cloned the repository (see Prerequisites above)
2. ✅ WhisperX service running at http://localhost:48001
3. ✅ SoX installed on your system

Then press Cmd+Shift+C to start using voice input!

Usage

Two Ways to Start Recording

Option 1: Keyboard Shortcut (Recommended)

• Press Cmd+Shift+C (Mac) or Ctrl+Shift+C (Windows/Linux)

Option 2: Status Bar Button

• Click the "Speak Claude" button in the status bar (bottom-right corner of VS Code)

Basic Usage

1. Start Recording: Use keyboard shortcut or click status bar button

   • Status bar will show "🎤 Recording..."

2. Speak: Say what you want to transcribe

3. Stop Recording: Press the shortcut again or click the status bar button

   • Extension sends audio to WhisperX
   • Transcribed text appears at cursor position

Using with Claude Code

1. Open Claude Code chat (Esc to focus input)
2. Start recording (keyboard or status bar button)
3. Speak your prompt
4. Stop recording
5. Text appears in Claude Code input

Configuration

Open VS Code settings (Cmd+,) and search for "Voice to Text":

• WhisperX URL: Default is http://localhost:48001
• Language: Language code (e.g., en, es). Leave empty for auto-detection.
• Diarization: Enable speaker labels (requires HF_TOKEN in WhisperX service)

Or edit .vscode/settings.json:

{
  "voiceToText.whisperxUrl": "http://localhost:48001",
  "voiceToText.language": "en",
  "voiceToText.diarization": false
}

Troubleshooting

"sox not found"

Install SoX (see Prerequisites above).

"WhisperX service is not running"

Start the WhisperX service:

cd whisperx-service
uvicorn main:app --port 48001

"No speech detected"

• Speak closer to the microphone
• Check system microphone permissions
• Ensure microphone is not muted

"Transcription failed: timeout"

The audio file might be too long. Try shorter recordings (under 30 seconds).

How It Works

1. Recording: Uses SoX to capture audio from your default microphone
2. File: Saves as 16kHz mono WAV file (optimized for speech)
3. Upload: Sends to WhisperX /transcribe endpoint
4. Response: Gets transcript text
5. Insertion: Inserts text at cursor position or copies to clipboard

Keyboard Shortcuts

Commands

Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P):

• Voice to Text: Start Recording - Start/stop voice recording

Requirements

• VS Code 1.85.0 or higher
• Node.js (for development)
• SoX (for audio recording)
• WhisperX service running on localhost:48001

License

MIT

Credits

Built for the MeetingMind AI project. Uses WhisperX for transcription.