Whisper WebGPU (TypeScript)

Real-time in-browser speech recognition using Whisper and WebGPU, written in TypeScript.

This is a TypeScript version of the original webgpu-whisper example, with full type safety and better IDE support.

Features

• 🎙️ Real-time speech recognition
• 🌐 Runs entirely in the browser (no server required)
• ⚡ Powered by WebGPU for fast inference
• 🔒 Private - all processing happens locally
• 🌍 Multi-language support (99+ languages)
• 📝 Fully typed with TypeScript

Getting Started

Install dependencies

npm install

Run development server

npm run dev

Build for production

npm run build

Tech Stack

• Transformers.js for running Whisper models
• WebGPU for accelerated inference
• Vite for fast development and building
• Tailwind CSS for styling

How it works

1. The app loads the Whisper-base model (~200MB) from Hugging Face
2. It requests microphone access and starts recording audio
3. Audio is continuously processed and transcribed in real-time
4. All processing happens in a Web Worker to keep the UI responsive
5. Results are streamed back and displayed as they're generated

Browser Support

Requires a browser with WebGPU support:

• Chrome/Edge 113+
• Safari/iOS 16.4+ (experimental)

License

Same as the parent Transformers.js project.