Spoken Language Input for a Patient Note System

Abstract

In developing a system to help CTICU physicians write patient notes, we hypothesized that a spoken language interface for entering observations after physical examination would be more convenient for a physician than a more traditional menu-based system. We developed a prototype spoken language interface, allowing input of one type of information, with which we could experiment with factors impacting use of speech. In this paper, we report on a sequence of experiments where we asked physicians to use different interfaces, testing how such a system could be used as part of their workflow as well as its accuracy in different locations, with different levels of domain information. Our study shows that we can significantly improve accuracy with integration of patient specific and high coverage domain grammars.