The use of formal grammars in isolated word recognition

A method using finite-state grammars and nonrecursive context-free grammars is presented for the recognition of isolated words. Some form of 'training' is required to combat problems of variations in speech. In the training mode, one grammar for each word in the vocabulary is constructed directly from a set of sample strings of 'features' represented by symbols. In the recognition mode, an incoming string is analyzed to determine which grammar, if any, could have generated it. The word corresponding to such grammar is then said to have been recognized.