About

Mary Pietrowicz is a Teaching Assistant Professor in the Department of Biomedical and Translational Sciences at the Carle Illinois College of Medicine. She is also a Senior Research Scientist at the National Center for Supercomputing Applications, specializing in computational health, wellness, and creativity. Her research focuses on determining health and wellness states by analyzing human expression through speech, movement, writing, and interaction, with an emphasis on creating scalable, deployable systems and tools. Her recent work explores speech and language models for conditions such as amyotrophic lateral sclerosis (ALS), depression, anxiety, schizophrenia, and Huntington's Disease, utilizing signals like voice and movement collected via ubiquitous cameras and microphones. These approaches aim to extend the capacity of medicine and telemedicine, improve access to healthcare, and revolutionize current medical practices. Pietrowicz also investigates how elements of human expression function and interact, with the goal of leveraging models of human expression for automated health and wellness assessment. Her background includes work as a software engineer in academia and industry, contributing to projects in interactive multimodal art, indoor location tracking, tangible interfaces for education, collaboration systems, and smart environments. She holds a Ph.D. in Computer Science from the University of Illinois, a Master of Computer Science from Florida Atlantic University, and a Bachelor of Science in Electrical Engineering from Purdue University.

Research topics

• Computer science
• Speech recognition
• Psychology
• Medicine
• Audiology

Selected publications

• Communication Experiences of Neurological Voice Disorder Patients With Automatic Voice Assistants and Human Listeners

  Journal of Voice · 2026-03-01

  article
  PublisherDOI

• Patient Perceived Experience of Automated Voice Assistants in Oral Cavity Cancer

  Otolaryngology · 2026-03-10

  article

  OBJECTIVE: Effective communication with humans and automated voice assistants (AVA) is essential for full participation in modern life. Because AVA relies on intelligible speech, individuals with speech impairments, such as those treated for oral cavity cancer (OCC), may face barriers when interacting with AVA. We assessed self-reported communication difficulty in patients treated for OCC related to AVA and humans. STUDY DESIGN: Prospective cross-sectional cohort study. SETTING: Out-patient clinics in two tertiary care centers. METHODS: This study was conducted between July 2024 and March 2025 and included adults (≥18 years) diagnosed with and surgically treated for OCC (experimental group) and those without speech impairments (controls). Participants completed the Speech Communication Experience Survey (SCES), which assessed perceived intelligibility when communicating with familiar partners, strangers, and AVA; and, as reported elsewhere, the Speech Handicap Index (SHI). RESULTS: In total, 159 participants were included (69 OCC, 90 controls). OCC participants were more likely to report that familiar partners, strangers, and AVA "almost always" or "always" had difficulty understanding their speech (P < .05). Free flap reconstruction and maxillectomy were associated with greater reported communication difficulty. AVA usage frequency and overall opinions toward AVA did not significantly differ between groups. CONCLUSION: Patients treated for OCC report significantly greater difficulty being understood by both AVA and human listeners compared to controls. These perceptions may inform peri-operative counseling and guide the development of more inclusive AVA technologies capable of accommodating diverse speakers.

  PublisherDOI

• The Effect of the performance of computer components and the lighting used in the scene on the speed and quality of the generated image in Blender

  Sustainable Production Instrumentation and Engineering Sciences · 2025-09-08

  articleOpen access

  One of the most common and popular programs for creating 3D graphics is the free and still under development Blender software. It has two main scene rendering engines - "Cycles" and "EEVEE"; both offer different effects and meet different user needs. The main features of each are mainly greater realism when producing scenes in Cycles, accuracy in light reflections and attention to volumetric effects. EEVEE, unlike its predecessor, is not physically correct, it shows the image in real time, so it generates scenes incredibly fast, but never realistic. In this article, a comparative study of the speed of the two rendering engines was carried out according to different workstations with other hardware components. For this, a simple scene with several light sources was built and a number of renders were made.

  PublisherOA PDFDOI

• Automated acoustic voice screening techniques for comorbid depression and anxiety disorders

  JASA Express Letters · 2025-02-01

  articleOpen access1st authorCorresponding

  Anxiety disorders (AD) and major depressive disorders (MDD) are growing in prevalence, yet many people suffering from these disorders remain undiagnosed due to known perceptual, attitudinal, and structural barriers. Methods, tools, and technologies that can overcome these barriers and improve screening rates are needed. Tools based on automated analysis of acoustic voice could help bridge this gap. Comorbid AD/MDD presents additional challenges since some effects of AD and MDD oppose one another. Here, acoustic models that use acoustic and phonemic data from verbal fluency tests to discern the presence of comorbid AD/MDD are presented, with the best results of F1 = 0.83.

  PublisherDOI

• Exploring Automated Detection of Barretts Esophagus via Machine Modeling and Acoustic Analysis

  2024-01-17

  articleOpen access1st authorCorresponding

  Gastroesophageal reflux disease (GERD) affects approximately 18-27% of adults in North America; and chronic GERD is associated with Barrett's esophagus (BE), a precursor to esophageal adenocarcinoma.Current screening and diagnostic procedures for GERD/BE are invasive, expensive, and uncomfortable for the patient.Automated screening tools for GERD/BE based on voice analysis and modern machine learning techniques could, however, potentially enable early detection of GERD/BE without invasive procedures.In this study, standardized, scripted speech is collected, analyzed, and compared across three groups, including a) patients with BE (BE+), b) patients without endoscopic evidence of BE (BE-), and c) patients without GERD and without voice symptoms (normal).Acoustic differences across groups are reported.In addition, multiple machine learning techniques are explored, and machine models are trained to detect the BE+ condition.The ability of selected machine learning models to discern across BE+, BE-, and normal conditions is reported.

  PublisherOA PDFDOI

• Vocal Effort and Acoustic Analysis of Gargle Phonation Versus Water Swallow in Patients With Muscle Tension Dysphonia: A Clinical Trial

  Journal of Voice · 2024-03-01 · 1 citations

  article
  PublisherDOI

• Developing a Machine-Learning Model for Detecting Intelligibility Differences in Individuals with Voice Disorders: A Feasibility Study

  2024-01-17

  articleOpen access1st authorCorresponding

  Voice disorders can reduce an individual's ability to produce intelligible speech; however, intelligibility in dysphonia has limited study.Current methods of intelligibility assessment are subjective and timeconsuming, making reliable, efficient monitoring of patient progress difficult for clinicians.Machine-learning techniques, however, may provide novel, automated assessment solutions.This study aims to discover machinelearning models that differentiate habitual speech (HS) from hyperarticulated or "clear speech" (CS).Two corpora with same-subject recordings of HS and CS were used.The corpus consisted of 115 speakers, 65 healthy and 50 with mild-to-moderate voice disorders, saying six sentences from the Consensus of Auditory-Perceptual Evaluation.Acoustic analyses revealed significant differences between HS and CS in speech rate and CPP for female speakers.Various machine modeling techniques are explored for their ability to differentiate HS and CS, and the results are reported.

  PublisherOA PDFDOI

• Advanced Machine Learning Voice-Based Biomarkers for Characterization of Barrett's Esophagus

  Techniques and Innovations in Gastrointestinal Endoscopy · 2024-12-20

  article
  PublisherDOI

• Tu2002 ADVANCED MACHINE LEARNING VOICE-BASED BIOMARKERS FOR BARRETT'S ESOPHAGUS

  Gastroenterology · 2024-05-01

  article
  PublisherDOI

• Landmark-based analysis of speech differentiates conversational from clear speech in speakers with muscle tension dysphonia

  JASA Express Letters · 2023-05-01 · 1 citations

  articleOpen access

  This study evaluated the feasibility of differentiating conversational and clear speech produced by individuals with muscle tension dysphonia (MTD) using landmark-based analysis of speech (LMBAS). Thirty-four adult speakers with MTD recorded conversational and clear speech, with 27 of them able to produce clear speech. The recordings of these individuals were analyzed with the open-source LMBAS program, SpeechMark®, matlab Toolbox version 1.1.2. The results indicated that glottal landmarks, burst onset landmarks, and the duration between glottal landmarks differentiated conversational speech from clear speech. LMBAS shows potential as an approach for detecting the difference between conversational and clear speech in dysphonic individuals.

  PublisherOA PDFDOI

Frequent coauthors

• Diana M. Orbelo

  Mayo Clinic in Arizona

  13 shared

• Keiko Ishikawa

  Mayo Clinic in Arizona

  12 shared

• Karrie Karahalios

  12 shared

• Sara Charney

  Mayo Clinic Hospital

  10 shared

• Carla Agurto

  9 shared

• Guillermo Cecchi

  IBM (United States)

  9 shared

• Cadman L. Leggett

  8 shared

• Amrit K. Kamboj

  8 shared

Education

• PhD in Computer Science

  University of Illinois Urbana-Champaign

  2017

• Master in Computer Science

  Florida Atlantic University

  1994

• BS in Electrical Engineering

  Purdue University System

  1986

Awards & honors

• Modeling Expressive Laughter in Conversational Speech (2019)
• A New Approach for Automating Analysis of Responses on Verba…
• Analysis of ICU Patient Interviews (2018)
• Discovering Dimensions of Perceived Vocal Expression in Semi…
• Acoustic Correlates for Perceived Effort Levels in Male Scri…