[K99] Identifying Vocal Tract Gestures from Audio to Reimagine Biofeedback for Childhood Speech Sound Disorder
Ente: National Institute on Deafness and Other Communication Disorders
Scadenza: 2028-06-30
Importo max: 132.695 EUR
Paese: US
Descrizione
PROJECT SUMMARY/ABSTRACT
1 The number of American children with chronic speech sound disorder is frustratingly high,
2 because best-practice therapies involving intensity and biofeedback exist but are clinically rare.
3 The long-term goal of this work is to leverage artificial intelligence (AI) pathways to intense
4 biofeedback therapy, lowering the otherwise lifelong impact of chronic SSD without raising
5 clinician burden. While AI clinicians can effectively raise therapy intensity, they cannot yet
6 automate best-practice vocal tract gesture feedback. This project’s central hypothesis is that
7 automated verbal vocal tract gesture feedback and visual vocal tract gesture biofeedback for AI
8 clinicians is possible through acoustic-to-articulatory speech inversion, which requires only a
9 microphone for clinical use. The overall objectives of this proposal are to infer child vocal tract
10 gestures from audio with speech inversion (Aim 1), classify vocal tract gesture error subtypes to
11 predict verbal feedback (Aim 2), and personalize visual tongue biofeedback displays for vocal
12 tract gesture error correction (Aim 3). During the mentored K99 phase, Aim 1 uses supervised
13 clinical machine learning regression, testing the primary hypothesis that fine-tuning an adult
14 speech inversion neural network on child data improves gesture inference for unseen children
15 compared to training on adults or children alone. In the independent R00 phase, Aim 2 uses
16 supervised clinical machine learning classification, testing the primary hypothesis that
17 multimodal gestural-acoustic architectures improve classification of error subtypes compared to
18 gestures or acoustics alone. Aim 3 uses a 2x2 factorial design, testing the primary hypothesis
19 that personalized biofeedback displays, animated using speech inversion vocal tract gesture
20 estimates, will have greater fidelity to ground-truth tongue ultrasound biofeedback videos than
21 stock animations for individual error subtypes. This innovative research will reimagine
22 biofeedback, driving a paradigm shift that expands the use of best practice intense biofeedback
23 therapy in the clinic, over telepractice, and at-home. This project aligns with multiple NIDCD
24 strategic themes by harnessing advanced technology to improve treatment through an
25 implementation pathway that enables vocal tract gesture biofeedback to improve the lives of
26 people with speech-sound-based communication disorders. The mentored phase of this research
27 is supported by the top-tier academic environment at the University of Maryland, College Park.
28 This award provides critical training in speech kinematics, clinical machine learning, and speech
29 animation that builds on the researcher’s background in clinical practice, speech science, and
30 engineering to develop the next generation of multidisciplinary scientific leadership.
Istituzione: UNIV OF MARYLAND, COLLEGE PARK
PI: Nina R Benway
Progetto: 1K99DC023310-01A1
Settori: National Institute on Deafness and Other Communication Disorders
Vai al bando originale
Registrati gratis su Bandolo per trovare bandi compatibili con la tua azienda.