A Robust Musical Audio Search Method Based on Diagonal Dynamic Programming Matching of Self-Similarity Matrices

We propose a new musical audio search method based on audio signal matching that can cope with key and tempo variations. The method employs the self-similarity matrix of an audio signal to represent a key-invariant structure of musical audio. And, we use dynamic programming (DP) matching of self-similarity matrices to deal with time variations. However, conventional DP-based sequence matching methods cannot be directly applied for self-similarity matrices because they cannot treat gaps independently of other time frames. We resolve this problem by introducing “matched element indices,” which reflect the history of matching, to a DP-based sequence matching method. We performed experiments using musical audio signals. The results indicate that the proposed method improves the detection accuracy in comparison to that that obtained by two conventional methods, namely, DP matching with chromabased vector rotations and a simple matching of self-similarity feature vectors.