Online Music Performance Tracking Using Parallel Dynamic Time Warping

Abstract

Resource allocation is a critical issue in the implementation of a portable, low-latency and efficient system for interactive experience. To address this, we propose the parallel dynamic time warping (PDTW), that utilizes multi-thread computation on a modern multi-core system, for online audio-to-audio music alignment. We also discuss the evaluation methodology that benchmarks the trade-offs among latency, accuracy of alignment, and computing resource. The proposed system utilizes multiple DTW alignment processes and parallel computing technique to reduce the processing time, as well as to improve both the alignment accuracy and robustness to tempo variation. Evaluation on several datasets with artificial time stretching exhibits the capacity of the system in enriched concert experience.