AI-powered music separation model that splits songs into 4 stems (vocals, bass, drums, other) using hybrid Demucs4/MDX architectures.