A high-memory WebAssembly implementation of OpenAI's Whisper speech-to-text model running entirely in your browser.

Try it now! (Update this URL after deploying)

• ✅ Large File Support: Process audio files up to 30 minutes (vs 1 minute in low-memory builds)
• ✅ High Memory: 512MB initial, 2GB maximum memory allocation
• ✅ Multiple Formats: Support for .wav, .mp3, .mp4, .m4a, .ogg, .webm
• ✅ Download Results: Save transcriptions with timestamps
• ✅ Microphone Recording: Record directly from your microphone
• ✅ Multiple Models: Support for tiny, base, small models and quantized versions
• ✅ No Server Required: Runs entirely in your browser

1. Fork this repository
2. Enable GitHub Pages in repository settings
3. Visit your GitHub Pages URL
4. Start transcribing!

1. Clone and serve:

   git clone https://github.com/your-username/whisper-demo.git
   cd whisper-demo
   python -m http.server 8000

2. Visit: http://localhost:8000

3. Load a model: Click one of the model buttons (recommended: base.en)

4. Upload audio: Select your audio file (max 30 minutes)

5. Transcribe: Click "Transcribe" and wait for results

• index.html - Main demo page with enhanced UI
• main.js - High-memory WebAssembly build (1.4MB, official build)
• helpers.js - Utility functions
• coi-serviceworker.js - Required for SharedArrayBuffer support
• samples/jfk.wav - Sample audio file for testing

This demo uses the official high-memory build with:

• Memory Limits: 512MB initial, 2GB maximum
• Function Signature: 5-parameter Module.full_default()
• Audio Limit: 30 minutes (28.8M samples at 16kHz)
• Threading: 8 threads for faster processing
• WebAssembly: Single-file build with embedded WASM

This implementation matches the capabilities of the official demo at https://ggml.ai/whisper.cpp/ with the same memory limits and processing power, plus additional features like:

• Enhanced UI with better model management
• Download functionality for transcriptions
• Improved error handling
• Progress indicators

• ✅ Chrome/Chromium: Full support
• ✅ Firefox: Full support
• ✅ Safari: Full support
• ✅ Edge: Full support
• ⚠️ Mobile: Limited (large memory requirements)

Requirements:

• WASM SIMD support (available in all modern browsers)
• SharedArrayBuffer support (enabled automatically via service worker)

• First Load: May take 10-30 seconds to download and initialize model
• Processing Speed: ~2-3x real-time on modern computers
• Memory Usage: Up to 2GB for large files
• File Size Limits: 30 minutes max audio length

Want to modify or rebuild? See the original repository for source code and build instructions:

• Source: ggml-org/whisper.cpp
• Build guide: See examples/whisper.wasm/ in the source repo

This project uses the same license as whisper.cpp.

• OpenAI Whisper - Original model
• whisper.cpp - C++ implementation
• Emscripten - WebAssembly compilation