Current state-of-the-art machine learning models for speech analysis achieve impressive accuracy, but their approach does not align with mechanisms for human speech perception. My project will build on larger research to develop alignment models between acoustic cues and human speech, with two main tasks: classifying children who later develop reading deficits from their speech, and developing an automatic speech recognition system for assigning acoustic cues to human speech recordings. A base set of physical acoustic cues, which vary among speakers, can be mapped to abstract phonemes and then into the overall phrasal framework. Right now, these cues must be manually annotated, while an automated workflow would greatly increase the available training data for future research.

I am pursuing SuperUROP because I love applying linguistic understanding of human speech to computational problems, as well as learning how to utilize machine learning techniques ethically with a deep understanding of the source data rather than from a black box perspective. I am excited to dig into a longer term research project with more independence.