Are you emotional or depressed? Learning about your emotional state from your music using machine learning

Music plays an important role in our society and has applications broader than just entertainment and pleasure due to its social and physiological effects. There has been recent interest in music, and two active research topics are music information retrieval and music emotion recognition, where data mining and machine learning techniques are integrated with music features and annotations to extract music information such as genres, instrument and its emotional content. In this paper, a machine learning music perception model is proposed to identify emotional content of a given audio stream and study the emotional effects of music. In fact, our developed model has the capability to determine the emotional state of a region (e.g., city) that could be utilized in applications such as marketing, and many other facets of the society such as cognitive development, education, therapy and security. This emotion recognition task is performed by mapping musical acoustic features to corresponding arousal and valence emotion indexes using a linear regression model. A radio-induced emotion dataset (RIED) is compiled from the songs broadcasted on radio in four US major cities (i.e., Houston, New York, Los Angeles and Miami) between October 21, 2017, and November 21, 2017. The RIED is then used as input to the proposed perception model to observe the regional music emotion propensity.