A Complexity Feature Extraction Method by Chord Progression and Transition Density for Music Media Content.

This study presents a complex feature extraction method based on chord progression and transition density for music media content. While various metrics have been proposed to represent the features of music media content, viewing music media content as time series data and extracting complexity based on its temporal structure can provide a feature set that includes the temporal variations of music media content. In this study, we focus on chord features that are effective as features of music content as perceived by users. We define complexity based on three aspects: the frequency of chord changes, the individual complexity of chords, and whether chords are part of a diatonic chord. This method uses chord progressions, the time signature of each chord, the tempo (BPM) of the music, and the key of the music as inputs, thus enabling the extraction of the complexity of a musical composition. We assume that complexity is closely related to the preferences of human songs. These were compared in terms of complexity with respect to the notes that comprise the chord at a given time. We define these as global and local complexities because music media content is characterized by chords that change or do not change over time. This allowed us to represent an aspect of a user's musical preferences. By implementing this method, it is possible to compare the complexity of music from the perspective of the chord features. This allows for the construction of a system that selects music that matches a user's preference based on complexity. This research is verified through comparative experiments between different songs and different arrangements of the same song.