In this paper, we present a greedy EM (GEM) method for

training Gaussian mixture density (GMD) based acoustic models.

In the proposed approach, starting from a single Gaussian, GMD

is built up by sequentially adding new components. Each new

component is globally selected to avoid local optima. The

sequential procedure offers more control over the model

structure to achieve better coverage of data. GEM also provides a natural way of integrating information criterion for model complexity selection. Experimental results on WSJ task show that the new method performs consistently better than the conventional method in speech recognition word error rate.

PDF file

In  Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP'05)

Publisher  Institute of Electrical and Electronics Engineers, Inc.
© 2007 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.

Details