Abstract
This paper reports on the results of the application of a vector quantization technique to an isolated word recognition system. The basic idea underlying this system is to represent a speech spectral sequence by several discrete spectral symbols. In this system, word templates are represented as sequences of discrete phoneme-like (pseudo-phoneme) templates which are automatically generated from a training set of word utterances by a clustering technique. The new word recognition system and its advantages are explained. This recognition system is especially effective for speaker-dependent largevocabulary word recognition, as well as speaker-independent word recognition using multiple word templates.
