Continuous Music Emotion Recognition Using Selected Audio Features

In this article, we present system for dynamic dimensional music emotion recognition (MER) and comparison of the system efficiency depending on different audio features. Experiments were performed on the database provided within the Emotion in Music task of MediaEval2015 benchmark (EiMME2015). In addition to the baseline features that were defined by the EiMME2015 organizers and included in the development data, we selected two-dimensional cepstral coefficients (TDC), linear prediction (LP) based coefficients, and the group of musical descriptors to evaluate their discriminative properties for recognition of music emotion. The experiments show that a combination of the EiMME2015 baseline features, LP coefficients and proposed set of music features significantly increases system performance for both arousal and valence emotional dimensions.