Music Emotion Recognition Based on Term Frequency and Pattern Entropy

Ja-Hwung Su, Yi-Wen Liao, Liang-Yu Chen

Abstract: In fact, music emotion is a music preference factor because it can deliver the user’s implicit emotions while listening to music. Therefore, a number of researches have been made on how to effectively recognize the music emotions over the past few decades. So called music emotion recognition refers a set of algorithms learning from the relations between low-level features and high-level emotions for predicting the music emotions. Although these forerunners have proposed effective methods, the emotion-to-acoustic diversity limits the advancement. To aim at this issue, in this paper, we present a method to increase the recognition quality by clarifying the diverse relations. In the proposed method, the music is represented by patterns. Next, the emotion frequencies and pattern entropies are calculated to show the distribution between concepts and music. Once the distribution is derived, the music emotion can be predicted more successfully. To reveal the effectiveness of proposed method, a comprehensive experiment is conducted and the results show the proposed method is more promising than the state-of-the-arts methods in emotion recognition.