Skip to main content

Individual Differences in Dimensional Weighting in Speech Perception

Project overview

One assumption driving research on human perception is that different people solve the underlying computational problems in the same way. According to this assumption, the world supplies a set of constraints and statistical regularities to which humans adapt, and although some people might be better than others at detecting these patterns, the optimal strategy does not differ across individuals. This assumption, however, may not always hold. For example, for a listener who has difficulty processing a particular dimension the optimal strategy may be to direct attention away from it and towards other stimulus features.

One way to test this assumption is to present people with ambiguous stimuli that can be categorized in multiple ways to see whether individuals differ in how they are perceived. Two examples of this phenomenon that have recently piqued public interest, include an image which some people perceive as a white and gold dress and others perceive as blue and white and a spoken word which some listeners hear as 'Yanny' and others as 'Laurel'. Similarly, research on speech sound perception has shown that categorization of ambiguous sounds can vary across individuals, and that these individual differences are stable across testing sessions. These examples demonstrate that people differ in how they categorize perceptual objects, but the causes of these differences remain poorly understood.

The Birkbeck project lead for this project is Dr Adam Tierney and it has been funded by the Leverhulme Trust.

Research aims

  • Our first objective is to investigate what drives differences in how individuals categorize perceptual objects.

    We hypothesize that one reason why people do not agree on the identity of ambiguous stimuli is that there exist stable individual differences in the extent to which people rely on different sources of information. In particular, we propose that listeners place more importance on evidence drawn from dimensions which they are better able to process. For example, prior work has demonstrated large individual differences in the extent to which people can accurately encode auditory information such as pitch and duration, and strong difficulties in perceiving one of these dimensions can co-occur with preserved processing of the other.

    We hypothesize that listeners adjust their use of evidence accordingly, such that dimensions which are more accurately encoded are weighted more highly during perception. This hypothesis leads to several predictions. First, we will test whether individual differences in the importance listeners place on different sources of information are stable across different categorization tasks, including perception of speech and music. Second, we will test whether individual differences in the ability to process different types of auditory information predict which interpretation listeners will choose when faced with ambiguous stimuli.

  • Our second objective is to investigate the underlying neural and cognitive mechanisms which make possible these different perceptual interpretations.

    We hypothesize that listeners use selective attention to focus on aspects of sound which they are better able to process and ignore aspects of sound which they perceive less precisely. We will test two predictions which follow from this hypothesis:

    • First, we predict that when listeners are distracted by the presence of a competing talker, they will have less control over their attention, leaving them less able to adapt their speech perception strategies in response to changes in the input.
    • And second, we will use EEG to test the prediction that neural activity will be more strongly driven by auditory dimensions on which listeners place more importance during perceptual categorization.

Outputs and outcomes

This project will be the first systematic study of the source of individual differences in the strategies people use to categorize sounds. This marks a shift away from current theories of speech perception which focus on the 'ideal listener'. If, as we suggest, what is ideal for one listener is sub-optimal for another due to individual differences in the reliability of auditory perception, then models of speech perception will need to account for the idiosyncrasies of the individual listener.

The proposed studies are also the first to apply neural methods from prior research on selective attention to the study of dimensional weighting during auditory categorization. We suggest that listeners can compensate for difficulties with auditory perception by actively attending to more useful sources of information. If so, then this strategy may be less available in distracting environments, during cognitively demanding tasks, or in certain populations.