Randomized Trial of AI-Enhanced Maintenance Training With Visual-Acoustic Biofeedback
Trial Snapshot
- Phase
- Phase 2
- Status
- Recruiting
- Sponsor
- New York University
- Enrollment
- 76
- Locations
- 2
- Primary Endpoint
- 1. Percentage of "Correct" Ratings by Blinded Untrained Listeners for /r/ Sounds Produced in Word Probes
Study Overview
Brief Summary
Speech Sound Disorder (SSD) affects a significant portion of school-aged children, leading to social and emotional challenges that can persist into adolescence and adulthood. The number of productions necessary for a remediated speech sound to generalize to connected speech is challenging to achieve in practice, leading clinicians to call for accessible, reliable resources allowing children to continue therapy outside of direct clinical interactions. Artificial intelligence (AI) tools hold promise as a means to extend direct service delivery in speech-language pathology, but research investigating the topic has been limited. This study (Generalization with AI Navigation using staRt, or GAINS) will measure the effects of a course of AI-mediated home practice intended to promote generalization of gains made through biofeedback treatment in a related study, Visual-acoustic Intervention with Service Delivery In-person and Via Telepractice Trial (VISIT; NCT06517225).
Detailed Description
This study will measure the effects of AI-guided home practice as a follow-up to participation in a study of technology-enhanced treatment for speech sound disorder affecting the American English "r" sound. Participants will sign a single consent form that covers their participation in the original VISIT study (in which they are randomized to receive biofeedback treatment in person or via telepractice) and and the follow-up GAINS study (in which participants are randomized to a follow-up period of treatment as usual [TAU] or treatment as usual plus an AI-guided maintenance program with biofeedback [TAU+AI]).We will measure production accuracy at the start and end of GAINS to test our working hypothesis that children/adolescents with speech sound disorder will show higher accuracy in /r/ production after six weeks of TAU+AI than six weeks of TAU alone. All GAINS treatment sessions will be carried out in the home setting using online tools. Pre and post treatment evaluations will be carried out in the laboratory setting.
Study Design
- Study Type
- Interventional
- Allocation
- Randomized
- Intervention Model
- Parallel
- Primary Purpose
- Treatment
- Masking
- Single (Outcomes Assessor)
Masking Description
All perceptual ratings will be obtained from blinded, naive listeners recruited through online crowdsourcing. Following protocols refined in previous published research, binary rating responses will be aggregated over at least 9 unique listeners per token.
Eligibility Criteria
- Ages
- 9 Years to 17 Years (Child)
- Sex
- All
- Accepts Healthy Volunteers
- No
Inclusion Criteria
- •• Must be between 9;0 and 17;11 (years;months) old at the time of enrollment.
- •Must speak English as the dominant or equally dominant language
- •Must have begun learning English by age 3, per parent report.
- •Must hear a rhotic dialect of English from at least one speaker in the home if the home language is English.
- •Must pass a pure-tone hearing screening.
- •Must pass a brief examination of oral structure and function.
- •Must demonstrate age-appropriate receptive and expressive language abilities on the Clinical Evaluation of Language Fundamentals-5 (CELF-5).
- •Must have access to a laptop or desktop computer for study sessions.
- •Must have home wifi sufficient to support video calls in the event of randomization to the telepractice condition.
- •Must have completed participation in the VISIT randomized controlled trial.
- •Must score at least 60% correct in at least one treatment session during the related previous study, VISIT.
Exclusion Criteria
- •• Must not exhibit voice or fluency disorder of a severity judged likely to interfere with the ability to participate in study activities.
- •Must not currently have orthodontia that crosses the palate and cannot be removed.
- •Must not have history of permanent hearing loss.
- •Must not have an existing diagnosis of developmental disability such as cerebral palsy or Down Syndrome.
- •Must not have history of major brain injury, surgery, or stroke in the past year.
- •Must not have epilepsy with active seizure incidents with in the past 6 months.
- •Must not show clinically significant signs of apraxia of speech or dysarthria.
Arms & Interventions
AI-guided maintenance training with visual-acoustic biofeedback
Participants will receive visual-acoustic biofeedback treatment from a clinician in a private, password-protected WebRTC room.
Intervention: AI-guided maintenance training with visual-acoustic biofeedback (Behavioral)
AI-guided maintenance training with visual-acoustic biofeedback
Participants will receive visual-acoustic biofeedback treatment from a clinician in a private, password-protected WebRTC room.
Intervention: Treatment as Usual (TAU) (Behavioral)
Treatment as usual [TAU]
Regular community-based care.
Intervention: Treatment as Usual (TAU) (Behavioral)
Outcomes
Primary Outcomes
1. Percentage of "Correct" Ratings by Blinded Untrained Listeners for /r/ Sounds Produced in Word Probes
Time Frame: Immediately before the initiation of GAINS participation and again after the end of GAINS participation (6 weeks later)
To assess generalization of treatment gains to untreated words, participants will be assessed with standard probes (30 words \[considered the primary target\], 20 syllables, and 10 sentences containing /r/ in various phonetic contexts). Stimuli in each probe will be presented individually in randomized order with blocking by stimulus type (word, syllable, sentence). Individual words will be isolated from the audio record of each word probe and presented in randomized order for binary rating (correct/incorrect) by 9 untrained listeners who are blind to treatment condition and time point, but will see the written representation of each target word. We will use the proportion of "correct" ratings for each token as our primary measure of perceptually rated accuracy.
Secondary Outcomes
No secondary outcomes reported