Effects of speech training on the accuracy of speech recognition for an individual with a speech impairment

Abstract

The ability to train clients to use speech recognition systems more effectively enables a broad range of clients with speech impairments to benefit from this state-of-the-art technology. This study investigated the effects of speech training on the recognition accuracy of the IBM VoiceType (Version 1.00) speech recognition system. Isolated words and words in sentences were produced by a young man with mild dysarthric speech. The participant's recognition accuracy was determined by both VoiceType and 10 listeners. The number of sessions required for VoiceType's recognition accuracy to become stable was identified. The results showed that stability of recognition was reached earlier for words in sentences than for isolated words. Initial consonant-null and final consonant place contrasts were chosen for speech training because these contrasts were most frequently misrecognized by both VoiceType and the listeners. Speech training resulted in up to a 57% reduction in errors by VoiceType, suggesting that training is an effective method of improving speech recognition performance.