MUSDB18

MUSDB18 music source separation dataset to open original stem file (mp4), which is done internally you need stempeg library. Outcome of mp4 file is a 3 dimensional np_array [n_stems, n_samples, sample_rate]. firt dimension meanings: { 0: mixture. 1: drugs, 2: bass, 3: others, 4:vocals, } Original dataset is not cutted

Papers using MUSDB18 (8)

HAFM: Hierarchical Autoregressive Foundation Model for Music Accompaniment Generation2026MGE-LDM: Joint Latent Diffusion for Simultaneous Music Generation and Source Extraction2025Decoupling Magnitude and Phase Estimation with Deep ResUNet for Music Source Separation2021 · 38 citesKUIELab-MDX-Net: A Two-Stream Neural Network for Music Demixing2021 · 31 citesFew-Shot Musical Source Separation2022Multi-scale temporal-frequency attention for music source separation2022Spectral Mapping of Singing Voices: U-Net-Assisted Vocal Segmentation2024Improving Music Source Separation with Diffusion and Consistency Refinement2024