An algorithmic approach to music retrieval by emotion based on feature data

Emotions evoked in response to music are not easy to quantify. While the feature data for a particular music excerpt is concrete, the emotions felt by a person are subjective. Although personal experiences can influence a person's perception of music, the combination of music features, such as timbre, tonality and rhythm all contribute to the perceived emotion. In addition, humans often associate multiple emotions with a particular piece of music making it difficult to measure the emotions evoked by music using only a small subset of emotions. This research defines a set of algorithms to map detailed feature data extracted from a group of instrumental music excerpts to a complex combination of emotion categories. In this way, feature data can be used to represent emotion as a blend of moods, simulating a more authentic flow of human emotion.