A method for identifying repetition structure in musical audio based on time series prediction

摘要: This paper investigates techniques for determining the repetition structure of musical audio. In particular, we consider problem segment similarity from perspective time series prediction, where seek to quantify in terms pairwise predictability between segments. To this end, propose a novel approach based on multivariate modelling audio features. Using chroma and MFCC features assumption that correct boundaries have been previously obtained, evaluate proposed against Beatles dataset. We both Queen Mary Tampere University versions dataset annotations. obtain maximum F-score 84%. Compared randomised baseline approach, result corresponds performance improvement 26 percentage points.