Evaluation of text produced by a speech-to-text service

Abstract

An increasing number of Japanese deaf or hard of hearing students who will attend universities heightened the need for a PC-based speech-to-text service which provides the transcription of instructors’ speech into text. In our study seven Japanese university students evaluated to what extent the service reflects the original speech. This paper reports the factors that influence their evaluations and identifies the matters that demand special attention on the part of the service providers. It suggests that priority should be given to the subject of the sentence and keywords, followed by facts, attitude of the speaker, and the development of the sentence.