Accuracy of three speech recognition systems: Case study of dysarthric speech

Publisher Website
Google Scholar
Abstract
People who have dysarthric speech production and other motor disorders that limit their ability to access keyboards may benefit from speech recognition technology. This study compared the accuracy and improvement patterns of three speech recognition systems (Microsoft Dictation, Dragon NaturallySpeaking®, and VoicePad Platinum®) when presented with the speech of a traumatic brain injury survivor with mild dysarthria and a speaker without dysarthria. Data consisted of speech recognition accuracy percentages obtained following each of five training sessions using each system. Accuracy scores were based on participants' reading of preselected sentences and generation of novel sentences. Results showed that, for both speakers, the recognition accuracy of Dragon NaturallySpeaking was significantly better and more consistent than that of the other two systems. The pattern of performance across probes and across systems was consistent between speakers. Given the training procedures employed, all three systems performed with greater accuracy when used by the speaker without dysarthria than by the speaker with dysarthria. Considerations for use and selection of speech recognition software are presented.