Projects
Predicting Voice Disorder Status From Acoustic Measures
Voice disorders cause people to have atypical voice quality (i.e., hoarseness) and vocal fatigue, which makes it difficult to communicate. The most common types of voice disorders involve excessive muscle tension in the voice box. However, this is difficult to measure. One type of acoustic measure, called relative fundamental frequency (RFF), is thought to be sensitive to tension in the voice box. Another acoustic measure, called cepstral peak prominence (CPP), can detect atypical voice quality. I tested whether a combination of features related to RFF and CPP could predict whether a person had a voice disorder in a large sample of voice recordings. We found that the combination of measures had good predictive performance and worked better than either type of measure alone. This type of measure is accessible, inexpensive, and non-invasive and could be useful for screening people at high risk of voice disorders.
Sensorimotor Control of Speech
When we move our bodies, our brains make predictions about sensations associated with that movement, and we compare those predictions with our actual sensory feedback. If there is a mismatch between prediction and reality, we use that information to help learn and refine movements. For speech, the most important feedback is auditory. In the lab, we can create artificial mismatches by altering auditory feedback and playing it back to a person on headphones in near-real time while they speak (think autotune!). We use these experiments to learn more about how the brain controls movements for speech.
In one project, I looked at how reliable these experiments are when they are repeated over time in the same individuals. In another project, I measured how people with voice disorders respond to altered auditory feedback compared to people with typical voices, and measured changes in muscle activation around the voice box using surface electromyography (sEMG) when they hear altered feedback. I found that people with and without voice disorders had the same ability to make small volitional changes in the pitch and loudness of their voices, but showed differences in how they adapt implicitly to alterations in their auditory feedback. I also found preliminary evidence for a possible laryngeal stabilization response to altered auditory feedback for pitch. These studies are helping us to understand typical and atypical vocal motor control, why people develop voice disorders, and potential avenues for new methods of diagnosis and treatment.
Human Factors and User-Centered Outcomes Measurement
Measuring outcomes is important for research, design, clinical practice – basically any goal-oriented activity we do. When we measure outcomes, we should include the perspectives of different stakeholders. In doing so, we need to consider how their varied perspectives and experience might interact with our measurement tools. In addition, the design of the measurement tools themselves could impact how reliable and interpretable our outcomes measures are.
People with voice disorders may have atypical voice quality, such as rough, breathy, or strained-sounding voices. These differences can range from mild to severe. One way we measure the extent of voice quality differences is by listening to the voice and making ratings using standardized tools. Ratings may be made by people with different perspectives, such as trained clinicians, family members, and the patients themselves. To determine whether we can trust our measures, as well as how to interpret them, our key performance indicators (KPIs) are reliability (how similar ratings of the same voice are when repeated multiple times) and agreement (how similar ratings of the same voice are when made by different people). My research in this area has focused on identifying and quantifying the impact of human factors on voice quality ratings, such as listener experience. We have shown that clinical training improves reliability and agreement, but that clinicians and patients may focus on different things in their ratings. I have also conducted A/B testing of potential changes to measurement tool design that can improve our KPIs, even when ratings are compared between users with different experience.
Surveys are another important tool for measuring outcomes. In clinical settings, these are standardized questionnaires that quantify the impact of a health condition on a healthcare user’s life. Historically, these questionnaires were often created by “experts” such as healthcare providers and researchers, without consulting the user stakeholders. I have collaborated on projects using a combination of qualitative and quantitative methods to understand patient experiences and to validate a questionnaire for people with communication disorders. Crucially, this questionnaire was designed through an iterative process of interviewing stakeholders (patients) to identify themes and to evaluate, modify, and select proposed questions. In addition to widespread clinical use, it has proved invaluable for measuring healthcare users’ perspectives in clinical research.