HamNava: A Dataset for Multi‑Label Instrument Classification

Despite significant advancements in music information retrieval, much of the progress has focused on musical traditions rooted in Western cultures. One of the hindrances preventing researchers from delving further into other musical traditions is the lack of datasets. This work introduces a new dataset, HamNava, constructed for multi‑label instrument classification. The dataset consists of 6,000 audio excerpts from Iranian classical music with a length of five seconds, each fully labeled with the presence or absence of eight classical instruments and vocals by a flexible number of annotators. We detail the instrument selection process and the methodology used to crowd‑source the annotations. To encourage future work, we also provide statistical results, a dataset split, and a baseline cross‑cultural multi‑label instrument classification on the introduced dataset.