A Real-Time Beat Tracking System with Zero Latency and Enhanced Controllability

Abstract

Identifying beat positions in music recordings, a central task in Music Information Retrieval (MIR), is commonly referred to as beat tracking. Typically, this involves computing an activation function to reveal frame-wise beat likelihood and then conducting post-processing to derive final beat positions. Existing methods often operate offline, requiring access to the entire music track for processing. In this article, we introduce a real-time beat tracking system based on the predominant local pulse (PLP) concept, originally designed for offline use. Our main contribution is the successful transformation of the PLP-based algorithm into a real-time procedure. Unlike traditional offline methods providing static beat positions, our real-time approach dynamically captures changes in local pulse characteristics with each frame of an audio stream. This yields additional insights, including beat context, beat stability, and beat lookahead for predicting beats in advance. In this way, our system not only demonstrates high controllability for real-time applications but also can operate at zero latency. Additionally, we present experiments comparing our real-time beat tracking system with other models and evaluating the accuracy of our lookahead feature. Finally, we showcase two real-world applications for interactive music making and educational music gaming that creatively leverage our system's output. In summary, our real-time beat tracking system offers a lightweight algorithm that is particularly well-suited for interactive music software development.

Authors

• P. F. Meier
• Ching‐Yu Chiu
• Meinard Müller