Speech recognition software: an alternative to reduce ship control manning

Abstract

This study identifies factors affecting the performance of commercial-off-the-shelf speech recognition software (SRS) when used for ship control purposes. After a review of research in the feasibility and acceptability of SRS-based ship control, the paper examines the effects of: "A restricted vocabulary versus a large vocabulary," Low experience level conning officers versus high experience level conning officers, "Male versus female voices," Pre-test training on specific words versus no pre-test training. Controlled experimentation finds that: "The experience level of a conning officer has no significant impact on SRS performance." Female participants experienced more SRS errors than did their male counterparts. However, in this experiment, only a limited number of trials were available to assess a difference. "SRS with restricted vocabulary performs no better than SRS with large vocabularies." Using the software "correct as you go" feature may impact software performance. Following the user profile establishment, individual user training on two specific words reduces error rates significantly. This study concludes that SRS is a viable technology for ship control and merits further testing and evaluation.