Speech synthesis in background noise: effects of message formulation and visual information on the intelligibility of American English DECTalk.

The purpose of the current research was to investigate the intelligibility of synthesized speech in noise, when listeners are able to watch an individual using augmentative and alternative communication (AAC) formulate messages on-line and when they are listening to a speaker without any visual information. A total of 80 participants were randomly assigned to four groups, with 20 participants in each group. Each group listened to sentences delivered using a different message formulation strategy: prestored; audibly formulated (messages are formulated on-line and the listener is able to hear the formulation as the message is being encoded); audibly formulated with no repeat (the full sentence at the end is not repeated); and quietly formulated (the message is formulated on-line, but the listener is not able to hear the system feedback throughout the formulation). The speaker for this study was a 35-year-old woman with cerebral palsy who used a VOCA with DECTalk (Beautiful Betty, American English) to communicate. Half of the sentences were presented in an auditory-only condition and half were presented in an auditory-visual condition. The dependent variable was intelligibility, as measured by the percentage of words correctly transcribed by each listener. The overall intelligibility of the sentences in the Audibly Formulated with No Repeat group was statistically significantly lower than in each of the other message formulation type groups. Visual information did not have an effect on intelligibility for this speaker. Clinical implications, limitations, and directions for future research and development are discussed.

Re-designing scanning to reduce learning demands: the performance of typically developing 2-year-olds.

Children with severe motor impairments who cannot use direct selection are typically introduced to scanning as a means of accessing assistive technology. Unfortunately, it is difficult for young children to learn to scan because the design of current scanning techniques does not always make explicit the offer of items from the selection array; furthermore, it does not provide explicit feedback after activation of the switch to select the target item. In the current study, scanning was redesigned to reduce learning demands by making both the offer of items and the feedback upon selection more explicit through the use of animation realized through HTML and speech output with appropriate intonation. Twenty typically developing 2-year-olds without disabilities were randomly assigned to use either traditional scanning or enhanced scanning to select target items from an array of three items. The 2-year-olds did not learn to use traditional scanning across three sessions. Their performance in Session 3 did not differ from that in Session 1; they did not exceed chance levels of accuracy in either session (mean accuracy of 20% for Sessions 1 and 3). In contrast, the children in the enhanced scanning condition demonstrated improvements in accuracy across the three 10-20-min sessions (mean accuracies of 22 and 48% for Sessions 1 and 3, respectively). There were no reliable differences between the children's performances with the two scanning techniques for Session 1; however, by Session 3, the children were more than twice as accurate using the enhanced scanning technique compared to the traditional design. Results suggest that by redesigning scanning, we may be able to reduce some of the learning demands and thereby reduce some of the instructional time required for children to attain mastery. Clinical implications, limitations, and directions for future research and development are discussed.