Information theoretic methods for aligning audio signals using chromagram representations

摘要

We present in this paper an information theoretic approach for analyzing the chromagram at its finest temporal resolution and for finding precise alignments of the similar segments in a song. We introduce information theoretic similarity measures for scoring the similarity of the candidate repetitions and for providing a final interpretation of the overall structure. Finally, we present experimental results showing that the proposed approach succeeds in finding relevant repetitive parts in audio.