Overcoming unusability: Developing efficient strategies in speech recognition systems

Abstract

This paper describes changes in user error correction strategies over time in the use of large vocabulary desktop automatic speech recognition (ASR) systems. Users with minimal practice with such systems were found to have considerable difficulty with error correction [1,2]. Users with more extensive use were found to have improved overall performance compared to initial use subjects. This is attributed to development of multimodal strategies for error correction rather than to significantly improved speech recognition rates or use of speech-based error correction techniques. These results point to the importance of multimodal interaction in the acceptance of speech recognition technology. Copyright © 2012 ACM, Inc.