Sixth European Conference on Speech Communication and Technology

Budapest, Hungary
September 5-9, 1999

Parameter Tying and Gaussian Clustering for Faster, Better, and Smaller Speech Recognition

Ananth Sankar, Venkata Ramana Rao Gadde

Speech Technology and Research Laboratory, SRI International, Menlo Park, CA, USA

We present a new view of hidden Markov model (HMM) state ty­ing, showing that the accuracy of phonetically tied mixture (PTM) models is similar to, or better than, that of the more typical state­clustered HMM systems. The PTM models require fewer Gaussian distance computations during recognition, and can lead to recog­nition speedups. We describe a per­phone Gaussian clustering algorithm that automatically determines the number of Gaussians for each phone in the PTM model. Experimental results show that this method gives a substantial decrease in the number of Gaussians and a corresponding speedup with little degradation in accuracy. Finally, we study mixture weight thresholding algorithms to drastically decrease the number of mixture weights in the PTM model without degrading accuracy. More than a factor of 10 reduction in mixture weights is achieved with no degradation in performance.

Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Sankar, Ananth / Rao Gadde, Venkata Ramana (1999): "Parameter tying and gaussian clustering for faster, better, and smaller speech recognition", In EUROSPEECH'99, 1711-1714.