Sources & licenses
Vocal Remover credits & licenses
What the Vocal Remover is built with, and the licenses that apply.
What we built
LoschaZ designed and wrote the Vocal Remover: the workspace, the audio decoding and playback, the spectrogram and segment processing around the model, the mixer and the WAV export. Everything runs in your browser.
Components & licenses
- Hybrid Transformer Demucs (htdemucs) by Meta Platforms, Inc., the source separation model: Demucs, MIT License.
- ONNX version of htdemucs used by this page (forward pass without the spectrogram stages): webnn/stem-separator, MIT License. Loaded from Hugging Face at a fixed revision and checked against its SHA-256.
- ONNX Runtime Web by Microsoft, the engine that runs the model with WebGPU or WebAssembly: ONNX Runtime Web, MIT License. Loaded from jsDelivr.
- Outfit and Inter, the typefaces of the site: SIL Open Font License 1.1 (Outfit, Inter).
Our changes
The model weights are used unchanged. LoschaZ wrote its own spectrogram, inverse spectrogram, normalisation and segment blending to match the original Demucs processing.