TITLE: ------ Hidden Neural Networks: A Framework for HMM/NN Hybrids AUTHORS: -------- S{\o}ren Riis Department of Mathematical Modelling, Building 305 Anders Krogh Center for Biological Sequence Ananlysis, Building 206 Technical University of Denmark DK-2800 Lyngby, Denmark Phones: + 45 4525 + ext. 3920,2470 Fax: + 45 45872599 emails: sr@imm.dtu.dk, krogh@cbs.dtu.dk ABSTRACT: --------- This paper presents a general framework for hybrids of Hidden Markov models (HMM) and neural networks (NN). In the new framework called Hidden Neural Networks (HNN) the usual HMM probability parameters are replaced by neural network outputs. To ensure a probabilistic interpretation the HNN is normalized globally as opposed to the local normalization enforced on parameters in standard HMMs. Furthermore, all parameters in the HNN are estim ated simultaneously according to the discriminative conditional maximum likelihood (CML) criterion. The HNNs show clear performance gains compared to standard HMMs on TIMIT continuous speech recognition benchmarks. On the task of recognizing five broad phoneme classes an accuracy of 84% is obtained compared to 76% for a standard HMM. Additionally, we report a preliminary result of 69% accuracy on the TIMIT 39 phoneme task. APPEARANCE: ----------- Accepted for 1997 International Conference on Acoustics, Speech and Signal Processing, ICASSP'97, Munic, April 21-24, 1997