A method for efficient storage and rapid application of context-sensitive phonological rules for automatic speech recognition

In an automatic speech-recognition system, the application of phonological rules to phonemic strings in order to create phonetic graphs is a computationally time- and storage-consuming process. A great many such graphs must be constructed during the decoding phase; thus it is important to be able to rapidly construct phonetic graphs for strings of words from graphs of individual words. However, because many phonological rules operate across word boundaries or require interword context, it is not possible to determine a unique, context-independent phonetic graph for a word. We describe a method for determining the phonetic graph for a word in isolation, together with auxiliary information to allow phonetic graphs for different words to be rapidly interconnected to form a phonetic graph for a string of words; the method also reduces storage requirements significantly.