5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Automatic Language Recognition Using High-Order HMMs

J. A. du Preez, D. M. Weber

University of Stellenbosch, South Africa

We present automatic language recognition results using high-order hidden Markov models (HMM) and the recently developed ORder rEDucing (ORED) and Fast Incremental Training (FIT) HMM algorithms. We demonstrate the efficiency and accuracy of pseudo-phoneme context and duration modelling mixed-order HMMs as well as fixed order HMMs over conventional approaches. For a two language problem, we show that a third-order FIT trained HMM gives a test set accuracy of 97.4% compared to 89.7% for a conventionally trained third-order HMM. A first-order model achieved 82.1% accuracy on the same problem.

Full Paper

Bibliographic reference.  Preez, J. A. du / Weber, D. M. (1998): "Automatic language recognition using high-order HMMs", In ICSLP-1998, paper 1074.