Stroke is devastating when left untreated. Early treatment significantly improves outcomes, and can be reliably detected using the FAST exam. The exam looks for facial asymmetry, arm weakness, and speech deficits as signs of stroke. Our research project aims to use smartphone technology to build a stroke detection algorithm based on the FAST exam. We have collected speech data of hospitalized stroke patients. The next step is to build an algorithm that can detect abnormal speech.

The speech data consists of both patient video and audio. Strokes can cause a number of different changes to a patient’s speech, including but not limited to aphasia, dysarthria, and paraphasic errors. Modeling speech will therefore need to take into account the wide range of potential deficits. Furthermore, the data collected is noisy, with extraneous input from the hospital. It will need to be pre-processed and isolated. This project will consist of two parts:

  1. Preprocessing and cleaning of data to isolate the patient’s voice.
  2. Modeling to detect specific categories of speech deficits as mentioned above.

Third-party API usage is limited to APIs that can be run locally because of the personal nature of the data.

This is an UNPAID research project.

Faculty Advisor

  • Professor: James Noble
  • Center/Lab:
  • This project is run under the joint mentorship of Drs. Noble and Williams, both professors in the Department of Neurology. Their past research together has investigated other novel approaches for early detection and intervention of stroke, including successful efforts such as Hip-Hop Stroke.

Project Timeline

  • Earliest starting date: 9/7/21
  • End date: 5/2/22
  • Number of hours per week of research expected during Fall 2021: ~10
  • Number of hours per week of research expected during Summer 2022: ~10

Candidate requirements

  • Skill sets: Machine learning
  • Student eligibility: freshman, sophomore, junior, senior, master’s
  • International students on F1 or J1 visa: eligible
  • Academic Credit Possible: Yes
  • Additional comments: Any past experience in speech recognition/processing would be a huge benefit, but is not a requirement