Query-by-conducting: An interface to retrieve classical-music interpretations by real-time tempo input

This paper presents an interface for finding interpretations of a user-specified music, Query-by-Conducting. In classical music, there are many interpretations to a particular piece, and finding "the" interpretation that matches the listener's taste allows a listener to further enjoy the piece. The critical issue in finding such an interpretation is the way or interface to allow the listener to listen through different interpretations. Our interface allows a user, by swinging a conducting hardware interface, to conduct the desired global tempo along the playback of a piece, at any time in the piece. The real-time conducting input by the user dynamically switches the interpretation being played back to the one closest to how the user is currently conducting. At the end of the piece, our interface ranks each interpretation according to how close the tempo of each interpretation was to the user input. At the core of our interface is an automated tempo estimation method based on audio-score alignment. We improve tempo estimation by requiring the audio-score alignment of different interpretations to be consistent with each other. We evaluate the tempo estimation method using a solo, chamber, and orchestral repertoire. The proposed tempo estimation decreases the error by as much as 0.94 times the original error.