Spoken emotion recognition using kernel discriminant locally linear embedding

Abstract

A new kernel-based manifold learning algorithm, called kernel discriminant locally linear embedding (KDLLE), is presented for spoken emotion recognition. KDLLE aims to make the interclass dissimilarity definitely larger than the intraclass dissimilarity in a reproducing kernel Hilbert space for the purpose of nonlinearly extracting the low-dimensional discriminant embedded data representations with striking performance improvement in spoken emotion recognition. Experimental results on the Berlin speech corpus demonstrate the effectiveness of KDLLE.