Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Discriminative MLPs In HMM-Based Recognition of Speech in Cellular Telephony

Sunil Sivadas (1), Pratibha Jain (1), Hynek Hermansky (1,2)

(1) Oregon Graduate Institute of Science and Technology, Portland, OR, USA
(2) International Computer Science Institute, Berkeley, CA, USA

Deviating from the conventional Hidden Markov Model-Multi-Layer Perceptron (HMM-MLP) hybrid paradigm of using MLP for classification, the proposed discriminative MLP technique uses MLP as a mapping module for feature extraction for conventional HMM-based systems. The MLP is discriminatively trained on the phonetically labeled training data to generate the phoneme posterior probabilities. We achieved a relative word error rate reduction of 15-35% on AURORA Phase 2 continuous digit recognition task defined by ETSI.

Full Paper

Bibliographic reference.  Sivadas, Sunil / Jain, Pratibha / Hermansky, Hynek (2000): "Discriminative MLPs in HMM-based recognition of speech in cellular telephony", In ICSLP-2000, vol.4, 153-156.