Analyzing speech and music blocks in radio channels: Lessons learned for playlist generation

Abstract

Customizing content according to preferences of the user and the current context is a key issue in electronic media. Audio content has some advantages over written text and video. Yet, apart from music playlists, little previous work has been performed on customizing audio content, i.e. speech-music playlist generation. The presented work makes a number of recommendations for speech-music playlist generation based on the program of broadcast radio channels. Nearly twenty thousand hours of audio content of twenty broadcast radio channels from four countries have been analyzed. Speech, music and mixed blocks were recognized automatically. The resulting data was analyzed for general statistics over the three types of audio blocks, for typical transitions and also for weekly and daily patterns. The paper also draws conclusions for customized speech-music playlist generation.