VOICE RECOGNITION APPARATUS

PURPOSE: To improve recognizing performance at the time of a turn of phrase by utilizing information showing that a candidate category is incorrect when a 1st spoken word is misrecognized for recognition processing for next pronunciation or subsequent processing, and removing the miscrecognition candidate category and processing the pronunciation. CONSTITUTION: Feature extracted data of a voice are applied to a recognition part 2 and a control part 1. The control part 1 measures the frequency of pronunciation of input data, receives an external NO (negative) indication, and indicates a category to be excluded to a dictionary control part 3 which controls a recognition processing command and the narrowing of dictionaries. A dictionary part 2b uses a dictionary to be matched for the recognition processing of a matching part 2a. Namely, when the recognition result of last pronunciation is misreconition, it is compared with an inputted recognition result and when the misrecognition category is included in candidate categories, the misrecognition category is excluded to determine a high-order candidate category. Consequently, it can be made easy to prevent an error of the same tendency and the recognizing performance by pronunciation of a turn of phrase can be improved.