Vocal/Music Classification using Incremental Learning Approach

Multimedia Data Content Labeling has been a research area for a relatively long period. Researchers have spent a substantial amount of time classifying video sequences, music genres, and artists' classification domains. vocal/music tagging is prevalent in this field, but it is addressed chiefly in Western music. This paper mainly focuses on the musicvocal binary classification of every 15-seconds window of the Indian songs, specifically in the Hindi language. The increasing amount of data recorded by the Hindi song industry shows that old methods are not efficient enough for tagging. Incremental learning has been used quite widely now in multimedia data content labeling. This paper proposes an incremental learning-based approach developed explicitly for binary classification based on Hindi songs. The proposed algorithm is developed for music-vocal classification. The incremental learning approach showed significantly good results. Moreover, there are not enough datasets available for performing such tasks for Hindi songs. Thus, this paper also introduced a novel dataset containing window-based information labeled for each song for each artist. The proposed method achieved remarkable accuracy of 83.6% for music vocal classification. Cross-Language Testing is also performed to check if the model generalizes over other language songs.