Theoretical Framework of A Computational Model of Auditory Memory for Music Emotion Recognition

The bag of frames (BOF) approach commonly used in music emotion recognition (MER) has several limitations. The semantic gap is believed to be responsible for the glass ceiling on the performance of BOF MER systems. However, there are hardly any alternative proposals to address it. In this article, we introduce the theoretical framework of a computational model of auditory memory that incorporates temporal information into MER systems. We advocate that the organization of auditory memory places time at the core of the link between musical meaning and musical emotions. The main goal is to motivate MER researchers to develop an improved class of systems capable of overcoming the limitations of the BOF approach and coping with the inherent complexity of musical emotions.