Speech recognition for older, pathological voices

Some diseases and brain injuries can seriously impair language. Patterns in an individual's speech can allow computers to describe these impairment with a high degree of accuracy. These techniques can be used to test large groups of people for drug trials and potentially replace pen-and-paper based testing methods. To fully automate this process, speech recognition systems can be used to automatically transcribe speech. Unfortunately, these technologies continue to perform relatively poorly for elderly speakers, or for individuals with speech disorders. Advancements in “deep learning” have the potential to create more powerful models and improve speech recognition accuracy for these groups. This project will explore methods for adapting speech recognition systems to these speakers to aid in impairment diagnosis.

Josh Ames
Faculty Supervisor: 
Alex Mihailidis
Partner University: