Title:  Intra-speaker Voice Quality Recognition for Voice Therapy

Committee:

Dr. David Anderson, ECE, Chair, Advisor

Dr. Elliot Moore, ECE

Dr. Christopher Rozell, ECE

Dr. Omer Inan, ECE

Dr. Eva van Leer, GSU

Abstract: A critical problem in voice therapy is poor extra-clinical adherence, stemming mainly from patients’ difficulty replicating and implementing their prescribed voice technique out- side of the therapy session. While clinicians can judge whether a patient’s voice quality resembles the individualized therapeutic target or not, patients have difficulty making this judgment themselves. The goal of therapy—replacing habitual voice production mechanics with optimal ones—cannot be achieved when patients cannot independently replicate the target voice technique and consistently differentiate it from their habitual voice production while speaking. Tools to help patients are lacking, demonstrating a substantial knowledge gap in clinical voice science. Machine learning methods have the potential to learn an in- dividual’s habitual and target voice qualities and subsequently classify future recordings accordingly. Classification results could serve as patient feedback in the clinician’s ab- sence. However, machine learning methods have primarily been applied to differentiate voice disorders or distinguish individual speakers from each other rather than identify in- dividual voice quality variations within a speaker. Therefore, this thesis aims to develop a tool that differentiates patients’ habitual voice quality from their target voice, thereby automating and extending the clinician’s judgment to the extra-clinical setting.