Using a connectionist network to eliminate redundancy from a phonetic lattice in a speech recognition system

In the Keal System (see G. Mercier, IEE Proceeding, vol.136, no.2 (1989)), as in most analytical speech recognition systems, the speech signal is processed in a bottom-up sequence of steps. In Keal, these steps are: acoustic-phonetic decoding, lexical analysis, and syntactic and semantic analysis. The purpose of this study is to simplify the result of the first module, namely the phonetic lattice, by removing its internal redundancy. This is achieved by using a connectionist network as a 'filter' in order to transform the lattice into a string. The task of the lexical decoder is simplified.<
>