Music Genre Classification Revisited: An In-Depth Examination Guided by Music Experts

Music Genres Reconsidered: Challenging Established Genres with a Data-driven Approach

“…Finally, the provided labels exhibit quality issues (e.g., replicas, mislabeling, and distortions) [24]. Pálmason et al [13] identified wrongly labeled examples in the ground truth using human raters (i.e., professional and semi-professional music protagonists) and updated them accordingly. However, these updates had only a minor impact on the classification error.…”

“…Second, areas in the vector space with a high density of artists with low proximity should correspond to the same genre. Third, not every artist should necessarily fall into a specific genre [13]. We refer to the cases that cannot be easily clustered as noise.…”

“…HDBSCAN scales well to many observations, automatically determines a suitable number of clusters, and treats outliers (i.e., ambiguous artists) as noise. The algorithm thus accounts for the difficulties with music genres [13]. We use the artist representations learned by CBOW, a 19-dimensional real vector normalized to unit length, as input.…”

“…Music genres allow consumers to easily find the music they are most likely to enjoy. However, music genres exhibit a significant shortcoming: Neither a ground truth nor a widely adopted standard exists [13,14]. As a result, almost every player in the industry uses a different number of genres.…”

Genre Classification of Traditional Malay Music Using Spectrogram Correlation

“…Some are culturally and historically derived, while newer labels may have been invented for marketing and commercial reasons. In this sense, music genres have no strict definitions and boundaries [1]. However, even with current labels, it is fairly obvious to the listeners that certain characteristics are shared within the members of a particular genre.…”