Estimating the Structural Segmentation of Popular Music Pieces Under Regularity Constraints

摘要: Music structure estimation has recently emerged as a central topic within the field of music information retrieval. Indeed, is highly structured stream, knowledge how piece organized represents key challenge to enhance management and exploitation large collections. This paper focuses on benefits that can be expected from regularity constraint structural segmentation popular pieces. Specifically, here, we study favors segments comparable size provides better conditioning boundary process. First, propose formulation task an optimization process, which separates contribution audio features one constraint. We illustrate corresponding cost function minimized using Viterbi algorithm. present briefly its implementation results in three systems designed for submitted MIREX 2010, 2011, 2012 evaluation campaigns. Then, explore efficient mean combining outputs selection presented at between 2010 2015, yielding level performance competitive state-of-the-art “MIREX10” dataset (100 J-Pop songs RWC database).