Event
Ph.D. Research Proposal: Saba Tabatabaee
Friday, August 28, 2026
10:00 a.m.
AVW 2168
Souad Nejjar
301 405 8135
snejjar@umd.edu
ANNOUNCEMENT: Ph.D. Research Proposal Exam
Name: Saba Tabatabaee
Committee:
Professor Carol Espy-Wilson, Chair
Professor Shihab Shamma
Professor Jonathan Simon
Date/time: Friday, August 28, 2026 at 10.00 AM
Location: AVW 2168
Title: Speaker Verification for Children in Classroom Environments and Towards a Noise-Robust and Language-Agnostic Speech Inversion
Abstract: This thesis develops robust speech processing systems across two areas: acoustic-to articulatory speech inversion (SI), which estimates articulator movements from the acoustic signal as vocal tract variables (TVs), and speaker verification for children's speech in real classroom environments.
The first part presents a unified SI system that jointly models oral TVs, nasal TV and glottal activity within a multi-task learning framework, where shared representations among related articulatory features improve prediction accuracy across all targets. This thesis also shows that the SI system generalizes beyond its English training data to other languages, demonstrating cross-lingual transfer and potential for multilingual applications in low-resource settings. Additionally, to enable the use of the SI system in noisy real-world environments, this thesis shows that combining data augmentation with joint training on a speech enhancement task improves noise robustness, benefiting both tasks. The SI framework is evaluated for clinical application in non-invasive nasalance estimation for children with velopharyngeal insufficiency (VPI), tracking velopharyngeal port movement over time to support clinical assessment and assist clinicians.
The second part of the thesis develops a speaker verification model for real classroom environments, where background noise and the natural co-occurrence of children's and adults' voices pose distinct challenges, and which can be used to provide teachers with a measure of student participation as an AI tool to support more effective learning environments.
