Real-Time Algorithms for Beat Tracking and Pitch Estimation

MIR is an interdisciplinary field focusing on computational methods for analyzing and interacting with music. A key application area in MIR is the development of interactive music systems that respond to musical input in real time, enabling new forms of expression in live performance, music education, and music-based gaming. These systems rely on real-time MIR techniques that must be causal, efficient, and robust, operating within strict timing constraints, on limited hardware, and under real-world conditions such as background noise and instrument cross-talk. This thesis focuses on two fundamental musical elements, rhythm and melody, and proposes real-time algorithms for beat tracking and pitch estimation tailored for interactive use. The first part of this thesis focuses on real-time beat tracking, which involves estimating the temporal structure of music by identifying beat positions, which is a fundamental component for synchronization and interaction in live musical settings. Our key contribution is the adaptation of PLP estimation from offline analysis to real-time use. The proposed algorithm operates with zero latency, continuously estimates beat positions, and provides stability metrics along with lookahead features for predicting upcoming beats. These enhancements enable the generation of pulse-based control signals that drive beat-synchronous audio effects, supporting dynamic interaction in live performance and mixing. The system is demonstrated in various practical contexts, including interactive music-making tools, rhythm-based educational games, and creative production environments. The second part of this thesis addresses real-time pitch estimation, focusing on tracking the fundamental frequency of musical signals in interactive settings. Our main contribution is RT-SWIPE, a real-time adaptation of the established SWIPE algorithm. This version is optimized for low-latency, multi-channel use and balances accuracy with computational efficiency to support robust pitch tracking in ensemble scenarios. To evaluate performance under real-time constraints, a delay-tolerant metric is introduced. This metric extends the standard RPA by accounting for algorithmic delays. The system is further examined in the context of pitch estimation for wind instruments, where cross-talk presents a common challenge. Based on these insights, a real-time intonation monitoring system is developed for wind ensembles. It provides immediate, individualized pitch feedback to musicians, supporting ensemble tuning and live musical communication. The system is tested in a real-world rehearsal environment, demonstrating its practical value. Finally, a music-based game prototype is presented, using real-time pitch input to enable interactive, creative gameplay. This showcases the broader potential of the system in educational and entertainment contexts. Through systematic evaluation, real-world testing, and application prototypes, this thesis advances real-time MIR with lightweight, responsive, and robust algorithms and systems that enhance musical performance, learning, and creative expression.