Evelyn Williams

Email: E.D.J.Williams@sms.ed.ac.uk

Research keywords: Self-supervised speech representations, human speech processing, linguistic structure, interpretability, algorithmic harm

Bio:

Evelyn is interested in how human brains and computers process speech and acquire linguistic structure.

She has a MSc in Speech and Language Processing from the University of Edinburgh, and an undergraduate MA in Linguistics from the University of Aberdeen. As an undergraduate student, she was awarded the James MacPherson Wattie Prize and was shortlisted for the LAGB Outstanding Undergraduate Dissertation in Linguistics award.

During her MSc studies she worked as a research assistant at the Centre for Speech Technology Research, focusing on building tools for evaluating text-to-speech and audio enhancement systems. After graduating, Evelyn worked on applications of machine learning for speech processing and NLP at startups – first as a Deep Learning Research Engineer at Audiostack, then as a Data Scientist at iovox. After a few years in industry, she came back to the University of Edinburgh to work on Open Research initiatives as part of the Research Data Support team.

Outside of work, Evelyn makes and sells pottery and likes to hang out at the beach.

PhD research:

Many state of the art speech technologies use self-supervised speech representations – that is, representations learned from unlabeled speech training data. Evelyn is interested in understanding the dynamics these representations use to encode linguistic structure, and whether we can learn anything about human language processing and acquisition from machine representations of speech. She is also interested in making speech models more interpretable and privacy-preserving for high-risk settings like healthcare.

Supervisors: Sharon Goldwater