On Beat Tracking and Tempo Estimation of Musical Audio Signals via Deep Learning

Automatic beat tracking and tempo estimation constitute important tasks in the Music Information Retrieval (MIR) field, with numerous relevant applications being developed. In this paper, a deep learning methodology for beat tracking and tempo estimation is proposed, ex-ploiting the Onset Strength Signal (OSS) of a musical audio signal. Specifically, the presented approach involves trans-forming the audio signal into its corresponding OSS and feeding it into a neural network to predict the occurrence of beats in time. Notably, the implemented system has a relatively simple structure and avoids higher complexity. The proposed method is evaluated on two benchmark datasets and compared to existing methods exhibits equiva-lent performance in terms of both beat tracking and tempo estimation accuracy.