Spontaneous continuous speech recognition system and method using mutiple pronunication dictionary

摘要

PURPOSE: A system and a method for sequentially recognizing a dialog voice using a multi-pronunciation dictionary are provided to enhance a recognition performance on the sequential dialog voice recognition, obtain a normalized output pattern, and enhance an efficiency of a language model by applying an extended pronunciation dictionary. CONSTITUTION: A sound model database(104) stores the information for a sound model. A language model database stores the information for the language model. A multi-pronunciation dictionary(105) database accepts various kinds of pronunciation variation information. A characteristic extractor(101) extracts/converts only the information useful for recognizing the inputted voice into a characteristic vector. A searcher(102) searches the highest probable word string through a Viterbi algorithm by using the sound model database and the language model database. A recognition result output tool(103) provides a recognition result for the inputted voice.