Senior Research Associate in Machine Learning for Speech Processing

About the Project

This Royal Society-funded project seeks to delve into the 'black box' of modern machine learning models that predict vocal tract shapes from audio recordings. The goal is to understand the intricate mapping between vocal tract movements and the acoustic speech signal. Utilizing state-of-the-art MRI recordings of the vocal tract during speech, the project aims to develop machine learning approaches that not only predict acoustic output but also reveal the underlying mechanisms. This requires hybrid machine learning (ML) approaches that integrate phonetic and physical domain knowledge with data-driven learning, alongside explainable AI (xAI) techniques to ensure model transparency and scientific validity. These approaches will be applied to a large database of real-time MRI and acoustic recordings. The outcomes will drive fundamental progress in critical applications like articulatory biofeedback for language learning and speech therapy.

Your Role...