VocalSplitter

Vocals Out, Music Stays.
جداسازی وکال از هر آهنگ، کاملاً داخل مرورگر شما.

Official codebase for VocalSplitter: Local Vocal Separation with HTDemucs in the Browser
Sajad Kazemi
Built with ONNX Runtime Web · WebGPU + WASM fallback

Search any song by how it sounds, not by what service you pay for.

🟢 build passing | 🟣 demo local-install | 🟢 status live | 💬 telegram join | 🟠 license MIT

What is it · Hear the difference · How it works · Install

🇮🇷 راهنمای فارسی: README-FA.md

What is it

Separate vocals from music locally in the browser with the HTDemucs model. No paid API, no subscription, no audio upload.

How it works

• Input via popup, resampled to 44.1 kHz
• Runs Demucs (demucs-web) with ONNX Runtime Web (WebGPU, fallback to WASM)
• Outputs *_vocals.wav + *_instrumental.wav
• First run downloads ~172MB model from Hugging Face and caches it in IndexedDB: https://huggingface.co/timcsy/demucs-web-onnx/resolve/main/htdemucs_embedded.onnx

Install in Chrome

1. npm install
2. npm run build
3. Go to chrome://extensions, enable Developer mode
4. Load unpacked → select the dist/ folder
5. Pin Vocal Splitter — Local and pick an audio file from the popup

Rebuild

npm install
npm run build

Limitations

• Heavy model: speed depends on CPU/GPU/RAM and track length. Keep the popup open until it finishes.
• Output is WAV (use Audacity/FFmpeg for MP3).
• May be imperfect on live recordings, heavy effects, choirs, or human-like instruments.

Privacy

Only the user-selected file is read. No access to history, cookies, or tabs. Network permission is only for downloading the public model.

Credits

Processing core from demucs-web (MIT) + ONNX Runtime Web. Check model license before public distribution.

---

جداکننده صدای خواننده — افزونه محلی کروم

جداسازی وکال از موزیک بدون API و بدون آپلود، با مدل HTDemucs داخل مرورگر. در اجرای اول مدل ~۱۷۲ مگابایتی از Hugging Face دانلود و در IndexedDB کش می‌شود.