Speech-language pathology ·
AI model grades dysarthria severity from real-world audio; speech recognition automates child language samples; biofeedback aids stroke swallowing
Listen to this post
read by AI voices · all episodesTHE SHORT READ
Singapore researchers reported in PLOS Digital Health that Whisper-ST1, a fine-tuned speech recognition model, graded dysarthria severity on real-world audio with agreement comparable to that between speech therapists. Other studies tested automated language sample analysis for developmental language disorder, sensor biofeedback for post-stroke swallowing, brain-computer interface displays for augmentative and alternative communication, smartwatch naming prompts in aphasia and app-delivered swallowing exercises.
THE DETAILS
- AI Oct 21, 2026Researchers from Synapxe and the National University Health System in Singapore fine-tuned a pretrained automatic speech recognition model into Whisper-ST1, a framework that classifies dysarthric speech severity from audio files, and reported the proof-of-concept study in PLOS Digital Health. Tested on real-world samples, 83 recordings from 12 speakers, it reached a per-speaker F1 of 79% (67% to 100%). In 66 samples from nine speakers, its agreement with two speech therapists (weighted kappa 0.814 and 0.878) was comparable to agreement between the therapists (0.851). The study had no independent external test set, and the authors called the findings preliminary.Source: PLOS Digital Health
- AI Oct 19, 2026A Journal of Speech, Language, and Hearing Research study ran narrative samples from 66 children with developmental language disorder (DLD) and 285 with typical language through automatic speech recognition and the proprietary LSA+ software. Speech recognition produced less accurate transcripts for children with DLD and for younger children, though automated measures correlated with manual ones (r = .79 to .98). Manual measures alone had better diagnostic accuracy than their automated counterparts, while composite models with multiple automated metrics yielded the highest accuracy.
- Research Oct 22, 2026In a randomized controlled trial in Archives of Physical Medicine and Rehabilitation, 30 patients with post-stroke dysphagia received conventional swallowing therapy including supraglottic swallow training, and 15 also received bimodal biofeedback combining an accelerometer and a nasal airflow sensor during the maneuver, 30 minutes a day, six days a week, for two weeks. Functional Oral Intake Scale scores were significantly higher with biofeedback after treatment and at one-month follow-up (P < .001), and the share of Penetration-Aspiration Scale scores of 6 or higher was significantly lower after the intervention (P = .026).
- Technology Oct 19, 2026Researchers tested whether photographic visual scene displays can be used with brain-computer interface (BCI) augmentative and alternative communication (AAC), which relies on neural signals such as the P300 event-related potential, reporting in the Journal of Speech, Language, and Hearing Research. Twenty-four neurotypical children aged 8 to 13 completed a contextual scene, an asymmetrical arrangement without context and a standard grid. Asymmetry increased P300 amplitudes, offline BCI performance was similar for scene and grid, and children rated the scene as more visually appealing. The authors called the results a preliminary foundation for pediatric BCI research.
- Technology Oct 23, 2026In a feasibility study in the American Journal of Speech-Language Pathology, 14 people with poststroke aphasia and 13 healthy controls completed a three-week, audio-based ecological momentary naming protocol delivered through smartwatch prompts while they went about daily routines. Average compliance was 77%, completion was 83% for controls and 84% for people with aphasia, and a one-item condition trended toward higher compliance (83.3%) than a nine-item condition. The authors said the protocol was feasible for adults with and without aphasia.
- Research Oct 20, 2026A cost-effectiveness analysis of the three-arm PRESTO trial, published in Head & Neck, compared app-based, therapist-led and paper-led prophylactic swallowing exercises in 150 patients receiving (chemo)radiotherapy for oropharyngeal head and neck cancer. Utilities did not differ significantly between groups. The app arm was dominated by the paper arm, with comparable utilities at higher cost, Therapist-led care, with 20 face-to-face sessions, had the highest costs, and its incremental cost-effectiveness ratio versus paper was €124,061.54 per quality-adjusted life year. The authors said daily therapist-led delivery was not cost-effective relative to paper-based self-management.Sources: Head & Neck, PubMed
The weekly innovation roundup for speech-language pathology, published every Sunday for the week before; every item links to its sources. Latest speech-language pathology news · All specialties