Recognition Of Multivariate Temporal Musical Gestures Using N-Dimensional Dynamic Time Warping

Abstract

This paper presents a novel algorithm that has been specifically designed for the recognition of multivariate temporal musical gestures. The algorithm is based on DynamicTime Warping and has been extended to classify any N dimensional signal, automatically compute a classificationthreshold to reject any data that is not a valid gesture andbe quickly trained with a low number of training examples.The algorithm is evaluated using a database of 10 temporalgestures performed by 10 participants achieving an averagecross-validation result of 99%.