Linear hidden transformations for adaptation of hybrid ANN/HMM models

This paper focuses on the adaptation of Automatic Speech Recognition systems using Hybrid models combining Artificial Neural Networks with Hidden Markov Models. A classical adaptation technique consists in adding a linear transformation network that acts as a pre-processor to the main network. We investigated the application of linear transformations not only to the input features, but also to the outputs of the internal layers. The motivation is that the outputs of an internal layer represent a projection of the input pattern into a space where it should be easier to learn the classification or transformation expected at the output of the network. 1 This paper is an extended version of the paper “Adaptation of Hybrid ANN/HMM Models Using Linear Hidden Transformations and Conservative Training” accepted for publication at ICASSP 2006.