The Effects of Errors in Speech Transcription: A User Study Misc

Tianyi Vera Bao, Afshan Ahmed, Wolfgang Stuerzlinger

Abstract:

This paper investigates how errors that occur during speech recognition affect users’ text entry performance. To study this, we implemented a speech recognition system that injects believable errors in a controlled manner. In our user study, participants were asked to transcribe a set of phrases using our speech recognition system, either with or without the insertion of errors. The results show that inducing 33% errors in a speech-based transcription task does not seem to affect users’ performance and experience in a significant manner. Yet, according to participants’ interview responses, our result might have been caused by the phrase set we used in the study. Our work thus motivates future research to develop a phrase set more suitable for speech-based transcription tasks.

Date of publication: Sep - 2022
Get Citation