Demucs

Demucs separates a finished song into vocals, drums, bass and the remaining accompaniment on your own computer. It's for musicians who need individual stems or a vocal-free backing track, and developers building audio tools. The project is archived and no longer maintained. Its Python code is open source under the MIT license.

The default Hybrid Transformer Demucs model combines waveform and spectrogram analysis, using a Transformer to connect the two representations of the audio. It also includes classic Hybrid Demucs and MDX models. A fine-tuned model trades longer processing time for potentially better separation, while quantized MDX models use less download and storage space with a possible loss in quality.

Demucs runs on Windows, macOS and Linux, with a community Docker option. It can process tracks on a CPU or use a CUDA GPU; typical GPU processing needs about 7 GB of GPU memory, though smaller audio segments reduce that requirement. Local processing keeps the audio on your machine. Google Colab and the Hugging Face Spaces demo offer cloud alternatives that process it remotely.

It accepts common audio formats such as WAV, MP3 and FLAC and exports separate stereo stems as WAV or MP3. Its karaoke mode produces vocals and accompaniment as two files. For people who prefer a graphical interface, Demucs-Gui and Ultimate Vocal Remover support Demucs, while a Python API lets developers include separation in their own applications.