Computerized speech recognition for anesthesia recordkeeping

Abstract

EARS, a computerized speech recognition system, has been developed to simplify data entry for automated anesthesia recordkeeping. As an isolated-word, speaker-dependent system, it uses a predefined vocabulary of about 350 words, each of which must first be pronounced by the user to train the computer to his voice. Thereafter, data are entered by spoken words separated by brief pauses and organized into sentences according to a specified syntax. Completed sentences are read back to the user by a speech synthesizer to verify the recognition. Errors can be corrected before they are recorded. Attention to human factors is an important design goal. Performance has not yet been measured, but appears acceptable.