Adapting Beat Tracking Models for Salsa Music: Establishing a Baseline with a novel dataset

This study addresses the challenge of adapting current beat tracking algorithms, predominantly trained on Western music, to the rhythmic complexities of Salsa, a genre rich in syncopations and polyrhythms. We benchmark the adaptability of three established models: BeatNet, Wavebeat, and Böck TCN, using our own newly introduced beat-annotated Salsa dataset and focusing on training methods that minimize the need for extensive annotated data. We find that, on Salsa music, models trained with popular datasets and fine-tuned with Salsa generally outperform models trained under other training conditions. This research not only establishes a baseline for beat tracking performance in Salsa music but also contributes to the broader goal of developing more universally adept music information retrieval systems.