STRUCTURAL ANALYSIS OF HARMONIC FEATURES USING STRING MATCHING TECHNIQUES

We focus on the harmonic content of music pieces, considering Harmonic Pitch Class Profiles as audio features to segment audio pieces. The features are compared using Optimal Transposition Index and correlation adapted measures, ensuring robustness to timbre variations or detuning issue s. No other audio feature is used for describing audio pieces. The identification of approximate repetitions is handled by string matching techniques applied to musical content. We define a new iterative inference algorithm detecting major repetitions, based on several existing string alignment al gorithms. The techniques employed ensure robustness to temporal changes or local transpositions between structur al parts.