Proceedings of the Annual Meeting of the Cognitive Science Society

Human perceptual grouping of sequential auditory cues hastraditionally been modeled using a mechanistic approach. Theproblem however is essentially one of source inference – aproblem that has recently been tackled using statisticalBayesian models in visual and auditory-visual modalities.Usually the models are restricted to performing inference overjust one or two possible sources, but human perceptualsystems have to deal with much more complex scenarios. Tocharacterize human perception we have developed a Bayesianinference model that allows an unlimited number of signalsources to be considered: it is general enough to allow anydiscrete sequential cues, from any modality. The model uses anon-parametric prior, hence increased complexity of thesignal does not necessitate more parameters. The model notonly determines the most likely number of sources, but alsospecifies the source that each signal is associated with. Themodel gives an excellent fit to data from an auditory streamsegregation experiment in which the pitch and presentationrate of pure tones determined the perceived number ofsources.