Speech emotion recognition based on supervised locally linear embedding

Speech emotion recognition is a new and challenging subject in signal processing area. In this paper, a new feature extraction method based on supervised locally linear embedding (SLLE) is proposed for speech emotion recognition. SLLE is used to implement nonlinear dimensionality reduction on high-dimensional emotional speech features with nonlinear manifold structure. And then the enhanced low-dimensional data representations embedded with SLLE are extracted for speech emotion recognition. Experimental results on natural emotional Chinese speech database confirm the validity and high performance of the proposed method.