Music Genre Classification and Feature Comparison using ML
Zhengxin Qi, Mohamed Rahouti, Mohammed Jasim, Nazli Siasi · 2022
An essential feature of the music is the genre, which can be considered a high-level description of an individual piece of music. In this sense, genre as a music feature is similar to typical descriptive features from the ML perspective. Although a genre can be understood as a principal component of a piece of music, the process of breaking it down to meaningful representation is a grand challenge. Identifying the genre with lower-level features is a key part of music genre recognition (MGR), which is an important field of research in music information retrieval (MIR). Understanding how to describe music genres in a quantitative way can be useful in analyzing the music for use in music recommender systems and the general understanding of music. This research aims to compare and analyze the feasibility, performance, and understandability of features used to describe music by predicting the genre using machine learning (ML) techniques. Using the mel-frequency cepstral coefficients (MFCC), a popular audio feature extraction method, key features from GTZAN, and human-understandable features from Spotify, this paper demonstrates a trade-off between classification accuracy, understandability, and interpretability of the features.