Development of a Remote Therapy Tool for Childhood Apraxia of Speech
Avinash Parnandi, Virendra Karappa, Tian Lan, Mostafa Shahin, Jacqueline McKechnie, Kirrie Ballard, Beena Ahmed, Ricardo Gutierrez-Osuna · 2015 · ACM Transactions on Accessible Computing (TACCESS) · doi:10.1145/2776895
Summary
This paper presents a multitier client-server system for remote administration of speech therapy to children with childhood apraxia of speech (CAS), a motor speech disorder where the brain has difficulty coordinating the movements needed for speech production. CAS requires intensive, frequent therapy — ideally multiple times per week — but access to qualified speech-language pathologists (SLPs) is limited by geographical barriers, SLP shortages, and family scheduling constraints. The system comprises three components: a tablet-based mobile application (iPad) that delivers speech production exercises to children in a game format, a back-end server with automated speech analysis modules that score the child's utterances, and a web interface that allows SLPs to remotely assign exercises, review recordings with automated scores, provide feedback, and adapt therapy programs. The speech analysis engine includes four modules: speaker diarization (distinguishing child from SLP speech at 74% accuracy), voice activity detection (96% accuracy for voicing delay, 94% for total production time), lexical stress classification (78% accuracy), and an HMM-based phoneme decoder (89% phone-level accuracy). The system follows the Nuffield Dyspraxia Programme 3 (NDP3) therapy protocol, supporting two therapy modes: SLP4 (therapist provides detailed knowledge-of-performance feedback after each attempt, four times per session) and SLP1 (therapist provides only knowledge-of-results feedback once at the end). A clinical pilot study with seven children (ages 4-7), their parents, and SLPs evaluated the system over a treatment period involving pretreatment baseline, treatment, and post-treatment assessments.
Key findings
The pilot study demonstrated feasibility and positive outcomes. Children in the SLP4 condition showed a median 70% improvement in speech sound accuracy from pretreatment to one week post-treatment, while the SLP1 group showed a median 40% improvement — consistent with the therapeutic expectation that more detailed clinician feedback produces better outcomes. Both groups performed similarly at one month post-treatment, suggesting the less intensive SLP1 mode may be sufficient for maintenance. Children responded better to sound production activities (emphasizing spatial accuracy of articulator movement) than stress production activities (emphasizing temporal control) in the short term, though stress gains continued improving to one month post-treatment. The system was well-received: 88% of respondents indicated they would use the tablet for therapy multiple times a week. Children found the app engaging and particularly enjoyed listening to their own recordings. Parents appreciated the convenience of in-home practice. However, participants requested more game-like features, animations, rewards, and audio-visual aids to maintain long-term engagement. SLPs valued the ability to remotely monitor progress and adapt exercises but noted concerns about ensuring remote therapy meets clinical standards.
Relevance
This research addresses a critical access gap in speech therapy services. CAS affects approximately 1-2 per 1,000 children and requires intensive therapy that many families cannot access due to SLP shortages, particularly in rural and underserved areas. By enabling high-frequency home practice with automated scoring and remote SLP oversight, the system transforms the therapy model from entirely clinic-based to a hybrid approach that extends clinical reach without replacing the SLP's expertise. For accessibility practitioners, the system architecture — mobile client, automated analysis server, clinician web dashboard — is a reusable pattern for any remote therapy or rehabilitation application. The automated speech analysis modules demonstrate that while not yet matching human SLP accuracy, machine scoring can provide meaningful interim feedback and reduce the burden on clinicians. The finding that a less intensive feedback mode (SLP1) produced similar long-term outcomes to intensive feedback (SLP4) has significant implications for scaling speech therapy services, suggesting that automated systems need not replicate the full richness of face-to-face interaction to be therapeutically effective.
Tags: speech disorder · speech therapy · childhood apraxia of speech · remote therapy · mobile health · automated speech analysis · gamification · children