Abstract:

The campus radio station, WRCT (88.3FM), is in the process of deploying a software automation system for overnight broadcasting. It is imperative to maintain a consistent feel between selected songs over long spans of time. However, with only very rough classification data for any given song, what stops the computer from playing pop followed by death metal, since they're both "rock"? In this presentation and discussion session we'll look at the viability of retrieving information from preestablished human behavior (such as WRCT's music log for the past three years) and/or properties of raw audio data, in order to build a reference to assist with dynamic generation of radio playlists.