论文《FMA: A Dataset For Music Analysis》摘要: We present a new music dataset that can be used for several music analysis tasks. Our major goal is to go beyond the existing limitations of available music datasets, which are either the small size of datasets with raw audio tracks, the availability and legality of the music data, or the lack of meta-data for artists analysis or song ratings for recommender systems. Existing datasets such as GTZAN, TagATune, and Million Song suffer from the previous limitations. It is however essential to establish such benchmark datasets to advance the field of music analysis, like the ImageNet dataset which made possible the large success of deep learning techniques in computer vision. In this paper, we introduce the Free Music Archive (FMA) which contains 77,643 songs and 68 genres spanning 26.9 days of song listening and meta-data including artist name, song title, music genre, and track counts. For research purposes, we define two additional datasets from the original one: a small genre-balanced dataset of 4,000 song data and 10 genres compassing 33.3 hours of raw audio and a medium genre-unbalanced dataset of 14,511 data and 20 genres offering 5.1 days of track listening, both datasets come with meta-data and Echonest audio features. For all datasets, we provide a train-test splitting for future algorithms' comparisons.