Music structure analysis by subspace modeling

摘要: Automatic music structure analysis is casted as a subspace clustering problem. By assuming that the feature vectors extracted from specific segment are drawn single subspace, any sequence of such derived recording will lie in union many subspaces segments are. First, sparse and low-rank tested for by employing three types beat-synchronous audio sequences. Next, novel computational efficient method proposed, coined ridge representation (RRSC). The performance aforementioned methods assessed conducting experiments on manually annotated Beatles benchmark dataset. experimental results indicate that: 1) RRSC comparable or exceeds 2) outperforms state-of-the-art proposed analysis.