FMA: A Dataset For Music Analysis
This dataset addresses a bottleneck for researchers in MIR by offering a large, accessible resource for tasks like browsing and organizing music collections, though it is incremental as it builds on existing data collection efforts.
The authors tackled the limited availability of large audio datasets for music information retrieval (MIR) by introducing the Free Music Archive (FMA), providing 917 GiB of Creative Commons-licensed audio from over 106,000 tracks, and demonstrated its utility with baseline evaluations for genre recognition.
We introduce the Free Music Archive (FMA), an open and easily accessible dataset suitable for evaluating several tasks in MIR, a field concerned with browsing, searching, and organizing large music collections. The community's growing interest in feature and end-to-end learning is however restrained by the limited availability of large audio datasets. The FMA aims to overcome this hurdle by providing 917 GiB and 343 days of Creative Commons-licensed audio from 106,574 tracks from 16,341 artists and 14,854 albums, arranged in a hierarchical taxonomy of 161 genres. It provides full-length and high-quality audio, pre-computed features, together with track- and user-level metadata, tags, and free-form text such as biographies. We here describe the dataset and how it was created, propose a train/validation/test split and three subsets, discuss some suitable MIR tasks, and evaluate some baselines for genre recognition. Code, data, and usage examples are available at https://github.com/mdeff/fma