Mid-level Chord Transition Features for Musical Style Analysis

Abstract

Chords and their progressions are an important tonal music. Specifically, transitions between within a piece carry style-relevant information. formation from audio recordings, a naive approa tomatic chord estimation for computing chord la then derives transition statistics. Often, this is Markov Models involving the Viterbi decoding al since chords are often ambiguous, deciding on o sequence can be problematic, which heavily aff derivation of transition features. In this paper, mid-level features that capture chord transitions i method exploits the Baum–Welch algorithm, whi hard decisions on chord labels. Instead, we obtai tures that account for ambiguities among chord tions. In several experiments, we evaluate thes style classification scenario discriminating four h Western classical music. Our soft transition fe achieve higher accuracies than comparable hard thus demonstrating the descriptive power of the n