A speech recognition system for programs written in FORTRAN level is described. The acoustic analyser based on a psychological model for phoneme identification can represent the speech sound by a phoneme string which contains some acoustic features such as buzz and silence. Therefore a word, its corresponding item of the word dictionary, is readily expressed in Roman letters accompanied with some acoustic features. The linguistic processor, using global acoustic information in addition to linguistic one, can effectively restrict the number of candidate words. The recognition process has been executed for each block; here, a block is a part of speech uttered in a breath. A new method for matching the input symbol string with errors to the reference symbol strings is also described. The system contains 146 words including letters. The syntax is expressed by state transition diagrams and implemented using tables. It has a back-tracking function. The recognition score is about 91% in terms of blocks when FORTRAN programs are spoken by 3 adult males. Since we cannot yet classify some letters such as P and T, the system has some restrictions of use.

A speech recognition system for continuously spoken Japanese sentences SPEECH YAMANASHI is described. The system has a knowledge source, in which syntactic, semantic and vocabulary knowledge are contained. It can deal with not only simple sentences but also simple complex and compound sentences without any restriction regarding the tasks. Phoneme identification is performed in real time, and after a slight revision the speech sound is represented by a pseudo-phoneme string. In the syntactic knowledge system, all inflections of inflectional words are taken into account. The syntactic analyzer, using a syntactic state transition network, predicts following words and outputs their syntactic interpretations. The semantic knowledge system deals with the meanings of each word, in particular, it deals with case structures for verbs. The semantic knowledge is used for the prediction of following words and construction of a semantic network of input sentences. The systems with vocabularies of 254 and 538 words work for four Japanese fairy tales with 70 [86] and 45 [69] % sentence and block (enclosed by brackets) recognition scores, respectively; and if sentences are spoken phrase by phrase, sentence and phrase recognition scores of 73 and 93 % are obtained for the system with vocabulary of 1,021 words. The system has an additional version, which may be used as a speech input device of a word processor.

A large vocabulary (with 1019 words and 1382 kinds of inflectional endings) continuous speech recognition system with high predictability applicable to any task and have an unsupervised speaker adaptation capability is described. Phoneme identification is based on various features. Speaker adaptation is done using reliable identified phonemes. Using prosodic information, phrase boundaries are detected. The syntactic analyzer uses a syntactic state transition network and outputs syntactic interpretations. The semantic analyser deals with the meaning of each word, the dependency relationships between words, the extended case structures of predicates, associative function, in universally applicable forms. The extended case grammar with a set of four-items of the case structure and the dependency relationships between words are based on semantic attributes of relating words, and realizes, together with associative function, universally applicable high prediction capability.