Elliptical basis functions for segment modeling

Abstract

Until recently, state-of-the-art, large-vocabulary, continuous speech recognition has employed hidden Markov modeling (HMM) to model speech sounds. The authors previously (ICASSP-92 p.625-8) presented the concept of a segmental neural network (SNN) for phonetic modeling in continuous speech recognition and demonstrated that a feedforward neural network, used within a hybrid SNN/HMM system, is able to reduce by 20% the word error rate over the baseline HMM system. They describe two developments over the initial system. First, a novel way to generate fixed length segment representations based on the discrete cosine transform (DCT) is presented. Second, it is demonstrated that an elliptical basis function (EBF) network can be used in the same hybrid framework. >