International Symposium on Chinese Spoken Language Processing (ISCSLP 2002)

Taipei, Taiwan
August 23-24, 2002

Incorporating Probability into Support Vector Machine for Speaker Recognition

Tieyan Fu, Qixiu Hu, Guangyou Xu

Tsinghua University, Beijing, China

Support Vector Machines (SVMs) is basically a discriminative classifiers, while it is hopefully that incorporating probability into SVMs will achieve better performance. This paper briefly reviews some of the methods that can be used to carry out the combination. By following one of them, we make it suitable for the task of speaker recognition, and Gaussian Mixture Models (GMM) is used as the generative model to derive Fisher kernel. Preliminary experiments are performed on a speaker identification task. The results are compared with GMM and standard SVMs baseline systems, and some suggestions have been made for future direction.

Full Paper

Bibliographic reference.  Fu, Tieyan / Hu, Qixiu / Xu, Guangyou (2002): "Incorporating probability into support vector machine for speaker recognition", In ISCSLP 2002, paper 36.