Mary Woolliams

Finding your voice

Current solutions for enabling people who have lost their ability to produce speech are slow. Here, a neural decoder was designed that consisted of two parts: decoding of kinematic articulatory features from continuous neural activity in cortical speech areas, and using output from the first stage to decode acoustic features to generate artificial speech in real time. Sentences produced using this method could be accurately perceived by listeners and, crucially, representations of articulatory features were similar between participants, suggesting a single decoding system could be used for multiple individuals.