16th Annual Conference of the International Speech Communication Association

Dresden, Germany
September 6-10, 2015

Dimensionality Reduction for Speech Emotion Features by Multiscale Kernels

Xinzhou Xu (1), Jun Deng (1), Wenming Zheng (2), Li Zhao (2), Björn Schuller (3)

(1) Technische Universität München, Germany
(2) Southeast University, China
(3) Imperial College London, UK

To achieve efficient and compact low-dimensional features for speech emotion recognition, this paper proposes a novel feature reduction method using multiscale kernels in the framework of graph embedding. With Fisher discriminant embedding graph, multiscale Gaussian kernels are used in constructing optimal linear combination of Gram matrices for multiple kernel learning. To evaluate the proposed method, comprehensive experiments, using different public feature sets from the open-source toolbox openSMILE on various corpora, show that the proposed method achieves better performance compared with conventional linear dimensionality reduction methods and single-kernel methods.

Full Paper

Bibliographic reference.  Xu, Xinzhou / Deng, Jun / Zheng, Wenming / Zhao, Li / Schuller, Björn (2015): "Dimensionality reduction for speech emotion features by multiscale kernels", In INTERSPEECH-2015, 1532-1536.