Auditory streaming in cocktail parties and the extent of binaural benefit

Abstract

Studies that investigate the advantage of spatial separation of speakers in a cocktail party generally focus on one of two processing strategies. The first assumes a top-down mechanism in which the listener focuses attention on the known location of a target speaker. Glimpses of target speech are collected and combined to form an auditory stream. The second strategy makes use of interaural differences in perceptual input and exploits these in order to suppress interfering sounds. Equalization-Cancelation (EC) models typically follow this approach. In order to investigate the contributions of both mechanisms, a headphone experiment was conducted that explores auditory streaming based on binaural cues. Sequences of logatomes spoken by one target and two interfering speakers were presented. In this experiment a new type of stimuli was introduced in which the possibility to use binaural masking release cues was eliminated for each time-frequency interval (glimpse) while the localization cues of the dominating source were preserved. Thus, listeners could attend to spatially separated glimpses, but no EC processing was possible. The effect of the availability of masking release cues on successful streaming will be analyzed and discussed.