Evaluation of Speech Input Recognition Rate of AR-Based Drawing Application on Operation Monitor for Communication Support During Endoscopic Surgery.

In endoscopic surgery, the surgeon and the assistant use both hands to proceed with the surgical operation. Therefore, the excision site cannot be shown by hand from the image inside the body displayed on the endoscopic monitor. Since there is a lack of communication between the surgeon and the assistant, it is necessary to have a system that indicates the excision point and prevents discrepancies between the surgeon and the assistant. Therefore, we developed a communication system that conveys the excision site from the in-vivo image on the endoscope monitor by operating the head movement and speech input without releasing the hand from the surgical instrument. There was a problem in using speech input in a noisy environment, such as an operation site. In order to make the system generate as few errors as possible, it was necessary to use words with high recognition performance and few unintentional behaviors by mistakenly recognized voice commands. In this experiment, the performance of recognition in the operating room environment and the possibility of unintentional operation were evaluated for each syllable number of words. As a result, the high recognition rate was possible with commands of 3 to 7 syllables, and commands with four or fewer syllables may induce unintentional system behavior. Consequently, we proposed to use the words of 5–7 syllables, which were highly recognized and have few wrong recognitions for voice commands.