Modelling Speech Perception
Building a computational model of the behaviour of human beings is the ambition behind strong or cognitive artificial intelligence. Human speech perception, during the first year of life, adapts to the sounds of the inventory of the native language. How can we build computer models that, when exposed to recordings of speech in a language, learn to perceive speech sounds like humans do?
Early work in this domain used simplified, artificial data sets, but it has now become possible to apply machine learning techniques to raw speech corpora with the cognitive goal of modelling speech perception. An additional critical part of this enterprise is benchmarking. We do targeted psychological experiments eliciting speech perception behaviour we know to be particularly informative. We then test models that work directly from the signal up, putting them in the exact same experimental setting as a human participant.