Music-Inspired Texture Representation

Techniques for music recommendation are increasingly relying on hybrid representations to retrieve new and exciting music. A key component of these is musical content, with texture being the most widely used feature. Current techniques representing however inspired by speech, not music, therefore capturing correct nature texture. In this paper we investigate two parts well-established mel-frequency cepstral coefficients (MFCC) representation: resolution mel-frequencies related notes; how best describe shape Through contextualizing parts, their relationship a novel music-inspired representation developed. We evaluate applying it task recommendation. use build three models, based current state-of-the-art methods. Our results show that understanding representation, possible achieve significant improvement. This contribution will only improve content-based but allow systems take advantage stronger content component.