Music Thumbnailing by Hooks

Abstract

Music or audio thumbnailing is the procedure of finding a continuous fragment that can represent the whole musical piece. This study proposes to create thumbnails based on the perception of listeners to identify the most memorable and distin- guishable fragment. This aligns with the cognitive definition of hooks, the most catchiest part of a song. This study tested whether audio features previously used to define catchiness collude with representativeness. First, a user study was carried out to assign a score for representativeness and familiarity to fragments. There- after, audio features derived with the CATCHY toolbox were used to approximate these scores. The results indicate that features measuring intensity, commonality and recurrence influence representativeness positively. This matches previous results regarding catchiness. Additionally, familiarity did not seem to have an impact and no preferred segmentation method was found. Lastly, a new music thumbnailing method is proposed based on the features that could approximate representativeness the best.