Foscarin Francesco ; Schlüter Jan ; Widmer Gerhard

Abstract

We propose a system for tracking beats and downbeats with two objectives:generality across a diverse music range, and high accuracy. We achievegenerality by training on multiple datasets -- including solo instrumentrecordings, pieces with time signature changes, and classical music with hightempo variations -- and by removing the commonly used Dynamic Bayesian Network(DBN) postprocessing, which introduces constraints on the meter and tempo. Forhigh accuracy, among other improvements, we develop a loss function tolerant tosmall time shifts of annotations, and an architecture alternating convolutionswith transformers either over frequency or time. Our system surpasses thecurrent state of the art in F1 score despite using no DBN. However, it canstill fail, especially for difficult and underrepresented genres, and performsworse on continuity metrics, so we publish our model, code, and preprocesseddatasets, and invite others to beat this.