I am a clinician, acoustic phonetician, and engineer developing clinical speech technologies to make best-practice biofeedback available to those who need speech therapy.

I am a postdoctoral fellow in Electrical and Computer Engineering at the University of Maryland, College Park, working with Distinguished University Professor Carol Espy-Wilson in the Speech Communication Lab. My training and research are supported by an NIH NIDCD Pathway to Independence Award (K99/R00) and the American Speech-Language-Hearing Foundation.

Jump to: Profile | Research Themes

Profile

• Education/Training:
  • BA: Linguistics, Cornell University
  • MS: Communication Sciences and Disorders, The College of Saint Rose
  • PhD: Speech-Language Pathology/Neuroscience Track, Syracuse University
  • Postdoctoral: Electrical and Computer Engineering, University of Maryland
• Funding:
  • Principal Investigator: NIH NIDCD Pathway to Independence Award (K99/R00) 1K99DC023310; American Speech-Language-Hearing Foundation Speech Science Research Grant 16038; American Speech-Language-Hearing Foundation New Investigators Research Grant 14181.
  • Co-Investigator: NIH R01DC017476 (PI: Tara McAllister) and NIH R01DC020959 (PI: Jonathan Preston).
  • Previously funded as a trainee by NIH NIDCD T32DC000046 (PIs: Matt Goupell and Catherine Carr; mentor: Carol Espy-Wilson).
• Clinical Trials:
  • First author: ChainingAI (in progress); single-case randomization design comparing ultrasound and visual-acoustic biofeedback (registry).
  • Co-investigator or co-author: iChain, VISIT, and GAINS (in progress); C-RESULTS (primary outcomes, secondary outcomes); ultrasound biofeedback trial for childhood apraxia of speech.
• Translation:
  • Patent: Lead inventor, Clinical Speech Analysis System for Childhood Speech Disorders, U.S. Patent Application No. 18/599,837 (notice of allowance mailed June 2026).
  • Industry: Co-founder, President, and CEO, SpeechLink Incorporated.
• Research Methods:
  • Ultrasound and articulatory data processing: ultrasound data collection and analysis, electromagnetic articulography data collection and analysis, tongue contour tracking, pose estimation, vocal tract variable analysis.
  • Biofeedback and clinical methods: ultrasound biofeedback, visual-acoustic biofeedback, acoustic phonetics, single-case experimental and randomized controlled trial designs.
  • Machine learning and speech signal processing: PyTorch, supervised and transfer learning, foundational audio models, multitask learning, acoustic-to-articulatory speech inversion, forced alignment, speech recognition and synthesis.
  • Software engineering: Python and JavaScript, full-stack development, RESTful API design, automated testing, code review, Git, CI/CD.
  • Cloud and data: cloud deployment, relational databases, research data capture (REDCap), open-access corpus construction, Linux and cluster computing.
  • Statistics: R, SAS.

Research Themes

Extending Ultrasound and Articulography: Kinematic Speech Analysis from Audio

Ultrasound imaging and articulography track real-time movement of the tongue, which is normally hidden from view. But these methods are difficult to deploy outside specialized laboratories. My postdoctoral research develops clinically interpretable acoustic-to-articulatory speech inversion, a neural network method that estimates the tongue, lip, and jaw movements measured by ultrasound and articulography from audio alone. I have shown that inferred tract variables differentiate error subtypes in children with speech sound disorders and quantify coordination differences in childhood apraxia of speech. I have also shown that clinician ratings of speech correctness predict articulatory proximity to the target sound. My K99 builds on this work to estimate child vocal tract gestures, classify gesture errors, and personalize midsagittal tongue biofeedback displays for each learner.

Clinically Tested AI Clinicians for Speech Therapy

I develop neural networks for speech therapy, embed them in validated motor learning therapy paradigms, and test them in clinical trials. I am the lead inventor on PERCEPT, a neural network that predicts clinician judgment of speech correctness during therapy. With Jonathan Preston, I co-developed ChainingAI, the AI clinician that combines the PERCEPT neural network with the adaptive Speech Motor Chaining motor learning software for practice at home. A single-case experimental study showed improved /ɹ/ production, and a randomized controlled trial of ChainingAI is underway. I also work on responsible standards for developing and evaluating clinical AI, including an end-to-end overview of clinical speech AI.

Ultrasound Biofeedback and Acoustic Biofeedback for Speech Therapy

My colleagues and I have shown that biofeedback, including ultrasound and acoustic biofeedback, enhances motor learning in speech therapy, particularly when delivered intensively and even for children who did not respond to traditional therapy. With Tara McAllister, Jonathan Preston, and Elaine Hitchcock, I contributed to a large randomized controlled trial finding faster speech improvement with biofeedback than with motor-based treatment during acquisition. Ongoing trials test telepractice delivery and AI-mediated home practice. My K99 aims to deliver the benefit of ultrasound-style tongue displays from a microphone alone.

Speech Profiles of Speech Sound Disorders in Children

I have investigated how speech sound disorders manifest across articulatory, perceptual, and acoustic dimensions, and how to automate measurement of each dimension. I collaborate with Katie Hustad and Visar Berisha on automated measures of speech sound development that support personalized screening and assessment for pediatric speech disorders.

Child Speech Data and Research

I curated the PERCEPT Corpora, the largest open-access speech corpora for speech sound disorders in American English. I am the sole US-based co-founder of the Interspeech special sessions Connecting Speech-science and Speech-technology for Children’s Speech, and I helped introduce child speech topic areas to the main program.

Automated Language Sample Analysis

My colleagues and I have examined automatic speech recognition for clinical language sample analysis. Automated language sample metrics outperformed manual metrics in predicting language disorder status in children, reducing the burden of manual transcription and scoring.