Using voice suppression algorithms to improve beat tracking in the presence of highly predominant vocals

Beat tracking estimation from music signals becomes difficult in the presence of highly predominant vocals. We compare the performance of five state-of-the-art algorithms on two datasets, a generic annotated collection and a dataset comprised of song excerpts with highly predominant vocals. Then, we use seven state-of-the-art audio voice suppression techniques and a simple low pass filter to improve beat tracking estimations in the later case. Finally, we evaluate all the pairwise combinations between beat tracking and voice suppression methods. We confirm our hypothesis that voice suppression improves the mean performance of beat trackers for the predominant vocal collection.