Dartmouth Events

Abstract: At the intersection of computational audio research and augmented device research lies the potential for powerful, immersive human experiences. What if our augmentative devices could selectively enhance and mitigate sound sources in our environment as a function of our attention; generate personalized audio soundscapes that cater to our moods or emotions; or transport us effortlessly from the quiet of our bedrooms to the echoes of an ancient monument?  In this talk, I posit that the key to achieving these sorts of experiences is the development of machine learning models that treat both audio and human listeners as first-class citizens – models that consume and render audio as a cardinal modality while leveraging an understanding of human audition to improve performance.  I will discuss my research on three themes – (1) modeling how we perceive what we hear, (2) how our behavior reflects what we hear, and (3) how our environment shapes what we hear – and will cover topics such as automated gestalt annotations for sound objects, behavior-driven target speaker selection, and novel view acoustic synthesis.  More broadly, I will gesture towards a timely and emerging research field at the crossroads of machine learning, acoustics, and human perception that is rich with possibility for future study.

Bio: Ishwarya Ananthabhotla is a Research Scientist on the Meta Reality Labs Research Audio Team.  She works on machine learning applied to problems in room acoustics, spatial audio, auditory perception, and behavior and communication understanding in conversations. More generally, her research interests lie at the intersection of machine learning, audio, signal processing, and auditory cognition.  She completed her Bachelors in Electrical Engineering and Computer Science from MIT in 2015, her M.Eng. from MIT in 2016, and her PhD from the MIT Media Lab in 2021. She was supported by the NSF Graduate Research Fellowship (2016-2019) and the Apple AI/ML Fellowship (2020-2022), and spent time interning at Meta Reality Labs Research and Spotify Research.