Design of High Robust Voice Interface for Speech Activated Neurosurgical Workstation

Abstract

Several methods were developed in order to improve reliability and robustness of a voice interface designed for a speech activated neurosurgical workstation that is a command-number-limited system but directly involved in nurosurgical operations. We chose a commercial voice recognizer and synthesizer as basic voice environment of the voice interface. A parallel connection grammar structure in the command recognition and a practical operation procedure oriented logical structure in the command understanding were designed to remove misrecognition and misunderstanding. In the voice recognizing process, by means of setting noise pits and dynamically adjusting tolerance level, the substitution errors caused by uncorrected voice commands and noises in operating room could be greatly reduced. To guarantee the high robust voice control, we employed real-time voice feedback to confirm the accepted commands. Our testing results showed that employing these methods greatly improved voice interface robustness.