Abstract
Music genre classification is a fundamental task in music information retrieval, aimed at discerning the categorical placement, or genre, of a given musical piece. Such classification holds significant utility in facilitating music recommendations tailored to individual preferences. Traditionally, efforts in music genre classification have primarily concentrated on unimodal analyses, focusing either on textual (lyrics) or audio components of music. However, given that music communicates through multiple modalities, including lyrics, audio, and visuals, there exists a pressing need to integrate these diverse sources of information. Multimodal music genre classification thus emerges as a solution, leveraging the amalgamation of two or more modalities present in music production to discern genre categorizations effectively. Moreover, existing endeavors in music genre classification have predominantly centered on compositions in Western languages, with comparatively limited attention devoted to non-Western languages. This study addresses these gaps by employing deep learning and language models to undertake multimodal music genre classification specifically within the context of Sotho-Tswana musical videos. The findings of this study demonstrate the efficacy of multimodal music genre classification, showcasing superior performance compared to unimodal approaches relying solely on lyrics for genre classification.