Fast generation of audio signatures to describe iTunes libraries

Abstract

iTunes Signature Maker (iTSM) uses a feature-driven audio editing algorithm to rapidly generate a short sonic signature of an iTunes music library. iTSM stitches together small segments of songs, driving a concatenative algorithm with spectral features intrinsic to the audio files themselves and with environmental features which describe how those files have been used. This paper relates iTSM to projects in algorithmic playlist generation and music summarization and then describes the software's implementation in relation to the project's objectives: accessibility to a broad audience, accurate representation of users' musical preferences, and interesting and enjoyable results. The project is evaluated in terms of these objectives using both statistical data and anecdotal feedback collected from users.