888 - Nika Jurov / Adaptation to speech in noise through selective attention
888 - Nika Jurov / Adaptation to speech in noise through selective attention
December 6, Nika Jurov defends her 888 on "Adaptation to speech in noise through selective attention." Her abstract is below.
Adaptation to speech in noise through selective attention
Speech perception is an active process, where speech features are extracted from highly variable signal. Prior research has shown that listeners weight the features according to how reliable they are by their life-long listening statistics and that they can have multiple models, one per speaker or situation. However, new data shows that feature reliability also depends on specific environments and speakers. In addition, adaptation is very quick and therefore it may be more parsimonious to propose a single model rather than a variety of them. To show how listeners can flexibly up- or down-weight a feature according to how reliable they seem to be in a moment, I present a single model of this process. I use a multiple encoder beta variational auto encoder model to offer a mechanistic explanation of how listeners are able to flexibly adapt to different listening situations: using their dimension based selective attention to focus on the feature that is reliable in that moment, while down-weighting or almost ignoring the unreliable feature. This is done by linking together behavioral data, information theory inspired modeling approach and theoretical proposals of dimension based selective attention as a mechanism that is at play in dynamic adaptation.