TL;DR本文提出了一种名为 MATT 的多示例注意力机制,用于解决长尾数据情况下的音乐流派分类问题,实验结果表明该方法在大规模音乐流派基准数据集上显著优于其他最先进的基线方法。
Abstract
Imbalanced music genre classification is a crucial task in the Music
Information Retrieval (MIR) field for identifying the long-tail, data-poor
genre based on the related music audio segments, which is very prevalent in
real-world scenarios. Most of the existing models are designed for