Intensive Speech Motor Chaining Treatment and Artificial Intelligence Integration for Residual Speech Sound Disorders

Project Summary/Abstract Speech sound disorders impacting /ɹ, s, z/ may become chronic due to either ineffective or limited treat- ment. The long-term goal is to leverage theoretical and technological advancements to accelerate the develop- ment of accessible and effective treatments that mitigate reduced quality of life due to chronic residual speech sound disorders (RSSD). To this end, the validated motor-based RSSD treatment Speech Motor Chaining guides speech-language pathologists (SLPs) through high-fidelity, high-trial, rapidly adapting treatment by dosing and manipulating several principles of motor learning in real time. SLP-led Speech Motor Chaining has been effective for individuals whose errors persist after traditional treatment. However, at least two challenges remain: first, optimal treatment intensity is unknown. Second, SLPs need validated avenues for evidence-based practice when caseload size precludes optimal intensity. Therefore, the overall objective of this proposal is to optimize a suite of theoretically motivated, high-fidelity, motor-based treatments delivered at the appropriate intensity, despite practical barriers, for the sounds comprising 90% of RSSD: /ɹ, s, z/. The central working hypotheses, supported by our preliminary work, are that Speech Motor Chaining is (a) more efficacious when delivered intensively (i.e., closely spaced for a fixed number of sessions), and (b) also beneficial when practice is led by an artificial intelli- gence (AI) SLP. The theoretical rationale is that increasing intensity early in treatment will mitigate erred prac- tice between sessions, improving outcomes relative to more customary practice distributions, and that reliable AI-mediated practice is effective in the context of validated treatments. There are three aims: Aim 1: Deter- mine how intensive/distributed treatment affects speech sound learning in RSSD. A randomized controlled trial (n=84) will test the hypothesis that intensive SLP-led Speech Motor Chaining (i.e., bootcamp) leads to greater gains in speech sound accuracy compared to an equivalent number of customarily distributed sessions. Aim 2: Determine improvement in /ɹ/ production when Speech Motor Chaining practice trials are led by an Artificial Intelligence clinician. A multiple baseline single subject design will test the hypothesis that Chaining-AI, in which an AI SLP provides clinical feedback, facilitates clinically meaningful change in /ɹ/ production. Aim 3: Demonstrate breadth of clinical AI capability by optimizing mis- pronunciation classification algorithms for /s/ and /z/. Mispronunciation detection algorithms will be trained to recognize clinical speech errors affecting /s/ and /z/, replicating expert listener judgement with clini- cally-acceptable accuracy. This significant research addresses a critical need for theoretical/empirical guidance for treatment intensity, offering sorely needed recommendations in a system where ~6 million American adults have unresolved RSSD. This innovative research accelerates a paradigm shift in which combined SLP/AI service delivery could overcome barriers to effective, accessible, and sufficiently intensive treatment, for 90% of RSSD.