R. O. Araz

This work mine sampling annotations from a music database and split them for training and evaluation, finding that naively splitting the annotations places the same tracks in different sets, and constructs a graph from the annotations and split it over connected components.

A unified music identification system can achieve robust performance across track and version identification tasks with just 10 seconds of audio input.

Models trained on the new DiVers dataset show remarkable resilience to noisy and acoustically diverse inputs, outperforming traditional benchmarks.

OMAR-RQ sets a new open-source benchmark for music audio understanding, outperforming existing self-supervised models across a diverse set of music information retrieval tasks.