Content-based Music Information Retrieval System with Automatic DJ Mixing

We will demonstrate our content-based music information retrieval system with automatic DJ mixing, which proposes a new, entertaining way to listen to MIR results. Typical MIR systems present MIR results to the user in a list format, from which users select and play songs. In order to improve this dull user experience, our system not only conducts content-based MIR, but also automatically mixes the songs in the MIR results, such as human DJs do to entertain the audience. Our system is mainly composed of three processes: automatic beat/tempo extraction, content-based MIR, and automatic DJ mixing/playing. By the beat/tempo extraction process, beat/tempo information is applied to all songs in the database. The content-based MIR process retrieves songs from the database by means of contentbased similarity to the users’ query. Finally, the DJ mixing/playing process, adjusts the beat position and tempo of the songs in the MIR result list, to generate smooth transition between songs. In this mixing process, we have implemented a unique tempo adjustment technique, which can calculate the optimal tempo adjustment coefficient to minimize acoustic signal distortion. Our method has proved to be robust to beat extraction errors that are caused by tempo octave relationships.