Mood Dependent Music Generator

Music is one of the most expressive media to show and manipulate emotions, but there have been few studies on how to generate music connected to emotions. Such studies have always been shunned upon by musicians affirming that a machine cannot create expressive music, as it’s the composer’s and player’s experiences and emotions that get poured into the piece. At the same time another problem is that music is highly complicated (and subjective) and finding out which elements transmit certain emotions is not an easy task. This demo wants to show how the manipulation of a set of features can actually change the mood the music transmits, hopefully awakening an interest in this area of research. 1 Music Mood Taxonomy The first issue we had to deal with mood generation was the mood taxonomy. The set of adjectives that describe music mood and emotional response is immense and there is no accepted standard; for example Katayose et al.[4] use a set of adjectives including Gloomy, Serious, Pathetic and Urbane. Russell [9] proposed amodel of affect based on two bipolar dimensions: pleasantunpleasant and arousal-sleepy, theorizing that each affect word can be mapped in this bi-dimensional space by a combination of these two components. Thayer [10] applied Russel’s model to music using as dimensions stress and arousal ; although the name of the dimensions is different fromRussell’s their meaning is basically the same. Also, we find different names in different research while the semantic value stays the same. Since valence and arousal are the most commonly used terms in many affective computing research, we will use these terms in this paper. Thus the music is divided in four clusters: Anxious/Frantic (Low Valence, High Arousal), Depression (Low Valence, Low Arousal), Contentment (High Valence, Low Arousal) and Exuberance (High Valence, High Arousal). These four clusters have the advantage of being explicit and discriminable; also they are the basic music-induced moods (even if with different names) as discovered by Kreutz [5] (Happiness, Sadness, Desire and Unrest) and Lindstrom [6] (Joy, Sadness, Anxiety and Calm). 2 The Mood Modifying Features Following Liu et al.[7], our current system employs three factors as the features that characterize mood: Intensity, Timbre and Rhythm. Liu’s study was D. Reidsma, H. Katayose, and A. Nijholt (Eds.): ACE 2013, LNCS 8253, pp. 626–629, 2013. c © Springer International Publishing Switzerland 2013 Mood Dependent Music Generator 627 about mood information extraction, we applied the principles they worked with to instead generate music.