Utilizing AI to revolutionize perception training for children with residual speech sound disorder
Presentation Type
Abstract
Faculty Advisor
Elaine Hitchcock
Access Type
Event
Start Date
25-4-2025 9:00 AM
End Date
25-4-2025 9:59 AM
Description
When a speech sound disorder (SSD) extends past 8 years of age, these errors can be classified as a residual speech sound disorder (RSSD). Research has shown that production training alone may not be sufficient in resolving children’s residual speech errors and the addition of perceptual training may be necessary to build accurate representations of errored sounds. To date, perception research study outcomes have not readily translated to clinical application because of methodological challenges in creating accurate, natural-sounding, speaker-specific stimuli for children with RSSD who are unable to produce correct versions of errored phonemes. In this pilot study, 160 correct fricative productions were artificially generated from a single 10-year-old male participant with an /s/ distortion using an AI speech software program. Altered productions were combined with 213 naturally-produced /s/ stimuli from the same participant, included in a listening task for source differentiation (AI vs. natural) and accuracy ratings using blinded listeners. The project investigates the use of AI stimuli to create speaker-specific, personalized stimuli for perceptual training modules to enhance intervention outcomes. Preliminary ratings reveal an >80% correct rating match for AI stimuli based on group average. Average rater accuracy for sound source identification of AI stimuli was slightly greater than chance at 56%. Listeners’ inability to definitively discriminate AI stimuli from natural speaker specific stimuli provides preliminary justification for investigating the use of AI generated stimuli for auditory training as a clinical tool for improving perceptual accuracy of fricatives in children with RSSD.
Utilizing AI to revolutionize perception training for children with residual speech sound disorder
When a speech sound disorder (SSD) extends past 8 years of age, these errors can be classified as a residual speech sound disorder (RSSD). Research has shown that production training alone may not be sufficient in resolving children’s residual speech errors and the addition of perceptual training may be necessary to build accurate representations of errored sounds. To date, perception research study outcomes have not readily translated to clinical application because of methodological challenges in creating accurate, natural-sounding, speaker-specific stimuli for children with RSSD who are unable to produce correct versions of errored phonemes. In this pilot study, 160 correct fricative productions were artificially generated from a single 10-year-old male participant with an /s/ distortion using an AI speech software program. Altered productions were combined with 213 naturally-produced /s/ stimuli from the same participant, included in a listening task for source differentiation (AI vs. natural) and accuracy ratings using blinded listeners. The project investigates the use of AI stimuli to create speaker-specific, personalized stimuli for perceptual training modules to enhance intervention outcomes. Preliminary ratings reveal an >80% correct rating match for AI stimuli based on group average. Average rater accuracy for sound source identification of AI stimuli was slightly greater than chance at 56%. Listeners’ inability to definitively discriminate AI stimuli from natural speaker specific stimuli provides preliminary justification for investigating the use of AI generated stimuli for auditory training as a clinical tool for improving perceptual accuracy of fricatives in children with RSSD.
Comments
Poster presentation at the 2025 Student Research Symposium.