Quantifying the benefits of using an interactive decision support tool for creating musical accompaniment in a particular style

We present a human-centered experiment designed to measure the degree of support for creating musical accompaniment provided by an interactive composition decision- support system. We create an interactive system with visual and audio cues to assist users in the choosing of chords to craft an accompaniment in a desired style. We propose general measures for objectively evaluating the effectiveness and usability of such systems. We use melodies of existing songs by Radiohead as tests. Quantitative measures of musical distance - percentage correct and closely related chords, and average neo-Riemannian distance - compare the user-created accompaniment with the original, with and without decision support. Numbers of backward edits, unique chords explored, and repeated chord choices during composition help quantify composition behavior. We present experimental data from musicians and non-musicians. We observe that decision support reduces the time spent in composition, the number of revisions of earlier choices, redundant behavior such as repeated chord choices, and the gap between musicians' and non-musicians' work, without significantly limiting the range of users' choices.