Effectiveness of using discrete utterance speech recognition software

This descriptive study explored text generation speeds, recognition accuracy, and participants' perceptions of the advantages disadvantages of using discrete utterance speech recognition software. Six participants (ages 19 to 35) with physical disabilities and intelligible speech were interviewed about their experiences using their speech recognition software. Using this software on their home computers, the participants completed five dictation tasks. Average individual dictation speeds ranged from 9 to 15 words per minute and average recognition accuracy ranged from 62 to 84%. The use of formatting and correction commands resulted in an average of two utterances being required to generate each dictation word. Participants found that recognition accuracy was not acceptable and that their speech recognition software was appropriate for use with word processors but had limited use with other applications. This study found that discrete utterance speech recognition can be effective for people who cannot use a keyboard to write. However, the slow speeds of text generation achieved by the participants suggest that people who can use a keyboard to some extent (e.g., slow typists) may not be able to increase their speed by using discrete utterance speech recognition software. The advantages and disadvantages of discrete products that are also relevant to continuous products are discussed.