音声認識システムにおける単語辞書について

概要

論文の詳細を見る
Recently various kinds of speech recognition systems are being investigated and the authors also are developing recognition systems for both spoken FORTRAN programs and Japanese sentences. It is necessary for the system to be expansible : not only the specialist about speech recognition but also the nonprofessional should be able to expand the system. One of the problems on the system is in increasability of the number of words in the word dictionary. In this paper we are concerning to this problem. First, we present a generative method of each item of the dictionary, next, the method is applied to the dictionary in the speech recognition system for spoken FORTRAN programs and also in another speech recognition systems, and we discuss the separability among items in the dictionary. The generative method of dictionary includes two rules : i) transforming the pronunciation corresponding to Japanese letters into signs, and ii) rewriting the signs acording to the acoustic and phonetic knowledges. Under these rules, addition, rewriting and elimination of words are very easy. The dictionary made by the above mentioned rules is used in our speech recognition system and works fairly well in our experiments.
山梨大学の論文