Output details
11 - Computer Science and Informatics
Robert Gordon University
Music-inspired texture representation
This paper was accepted for oral presentation at this highly selective international AI conference (11.5% acceptance [AAAI Preface]). Its significance comes from the Mel-Frequency Spectrum (MFS) texture representation for music content that overcomes a speech bias in standard MFCC. A theoretical analysis of MFCC reveals loss of information at important parts of the note spectrum, addressed in the novel MFS. A rigorous evaluation using a large music collection demonstrates the advantage of MFS over several MFCC variants for music recommenders. Subsequently, QMUL’s Centre for Digital Music invited the authors to add an MFS plugin to QMVamp (www.vamp-plugins.org/download.html).