Repetition Coherence Reveals Computational Principles of Auditory Object Formation

Abstract

A foundational task for perceptual systems is to identify objects. In hearing, repetition -- for example of a feature critical to object identity -- is a powerful cue for object formation. The principles governing how repetition may underpin this process are not well understood, particularly for the noisy, approximate repetition typical of biological environments. To characterize the perceptual organization of acoustic scenes into objects, we adapted the very successful motion-coherence paradigm from vision research, developing a repetition-coherence framework to investigate the evidence accumulation that yields auditory perceptual objects. Participants heard dense tone-cloud sequences in which subsets of tones repeated across cycles. The proportion of repeating tones defined coherence, and the length of the repeating unit was varied. They performed two tasks: repetition detection, which captures the endpoint of object formation, and sensorimotor synchronization, which provides a continuous readout of object formation as it unfolds. The convergence between the two tasks validates sensorimotor synchronization as an online behavioral probe of an otherwise covert process. We discovered that at high coherence both detection and stable tracking were achieved at the same number of cycles, independent of other factors. Yet within this fixed integration regime, longer repeated units were less likely to give rise to a perceptual object, dissociating the likelihood of object emergence from its timing. This scale-independent integration limit suggests that auditory object formation is governed by two interacting principles: extraction of statistical regularities and a fixed integration window. The behavioral signatures parallel known neurophysiological dynamics, providing a helpful link to interpret them.