An empirical study on users' acceptance of speech recognition errors in text-messaging

Although speech recognition technology and voice synthesis systems have become readily available, recognition accuracy remain a serious problem in the design and implementation of voice-based user interfaces. Error correction becomes particularly difficult on mobile devices due to the limited system resources and constrained input methods. This research is aimed to investigate users' acceptance of speech recognition errors in mobile text messaging. Our results show that even though the audio presentation of the text messages does help users understand the speech recognition errors, users indicate low satisfaction when sending or receiving text messages with errors. Specifically, senders show significantly lower acceptance than the receivers due to the concerns of follow-up clarifications and the reflection of the sender's personality. We also find that different types of recognition errors greatly affect users' overall acceptance of the received message.