Spleeter

A highly popular, open-source audio source separation library developed by the music streaming platform Deezer. Built on top of the TensorFlow framework, it utilizes pre-trained convolutional neural networks (U-Nets) to analyze audio spectrograms rather than raw waveforms. It is engineered primarily for extreme speed and high throughput, capable of splitting audio tracks into two, four, or five distinct stems (including vocals, piano, and drums) up to 100 times faster than real-time playback. While it may introduce minor spectral artifacts compared to modern time-domain models, its low computational overhead makes it ideal for large-scale industrial processing, software integrations, karaoke app backends, and quick DJ sampling workflows.