A collaborative project involving speech pathologists, rehabilitation specialists, and technology developers from Australia, Singapore, and Pennsylvania produced a therapeutic Word Trainer Application that runs on an iPad, captures user confrontation naming performance via the built-in microphone, uses automatic speech recognition (ASR) software to judge utterance accuracy, and is designed to support client home use to improve naming performance. The present article reports results of the authors’ research to determine: (a) ASR-human agreement on word production accuracy; (b) observed therapeutic benefits to users upon program completion, and again at 1-month maintenance; and (c) levels of users’ satisfaction with app use.
Study subjects comprised five persons diagnosed with aphasia plus co-occurring apraxia of speech. The study’s word stimuli comprised 124 commonly-used nouns in 8 categories (e.g., animals, clothing, vegetables). From these, individualized 20-word target subsets were generated for subjects. In therapeutic use, the app displayed a picture of the target noun, and permitted levels of cueing including categorial identification, first letter identification, and audio display of first sounds to support verbal naming. ASR-human agreement of word production accuracy compared judgments of trained auditors against binary ‘accurate/inaccurate’ algorithmic determinations shown on screen. Multiple baseline designs were employed to establish client pre-intervention performance, assess improvements during use, and probe maintenance effects. At the end of the process, researchers asked clients about their satisfaction levels with their experiences in using the app.
During 4 weeks of home practice, with speech pathologists providing support remotely, subjects named a total of 7,069 targets, of which 5,346 were judged accurate by both ASR and speech pathologist judges, yielding an ASR-human agreement on word production accuracy of 75.7% (range: 65.1% – 82.8%). Engagement with the app varied by subject — log-ins/week ranged from 4.3 (Subj. 4) to 25.5 (Subj 3); and word productions per log-in ranged from 13.5 (Subj. 4) to 31.9 (Subj. 2). Single-subject, multiple baseline research showed that subjects improve over time with the ASR-based feedback, and that performance gains are largely maintained one month post-treatment. At the end, subjects reported enjoying using app when conjoined with remote support of a speech pathologist.
This exploratory study shows that – for these five subjects – performance in confrontation naming improved after supported home use of the Word Trainer Application, with gains significantly maintained after a month. The current work thus represents proof of concept of the approach and suggests value in further work to research app improvements, best candidates for improvement, protocols for clinical employments, and feasibility of reducing remote support by clinicians.
For further reading: K. J. Ballard, N. M. Etter, S. Shen et al., July 2019,
Feasibility of automatic speech recognition for providing feedback during tablet-based treatment for apraxia of speech plus aphasia. American Journal of Speech-Language Pathology, 28(2S): 818–834, https://doi.org/10.1044/2018_AJSLP-MSC18-18-0109