We often take for granted our ability to focus on what one person is saying against a background of other talkers and sound sources—the so-called cocktail party effect. Just how the auditory system processes the complicated sound signals reaching our ears remains a mystery. (See the Physics Today articles by Emily Myers, April 2017, page 34, and by Bill Hartmann, November 1999, page 24.) Angela Josupeit and Volker Hohmann of the University of Oldenburg now show that deciphering complex soundscapes doesn’t necessarily rely on complex auditory features; rather, sufficient information can come from sparse yet salient cues conveyed in the sound’s periodicity. In numerical simulations, the researchers consider time- and frequency-based analyses of four acoustic features, or “glimpses”: periodicity, related to pitch; periodic energy, related to the acoustic spectrum; and periodic timing and amplitude differences between the signals reaching each ear. The soundscapes comprised superposed recordings of two, three, or four people, virtually placed in separated locations. Each person said similar phrases that combined a “call sign,” a number, and a color; the signal-processing challenge was to identify and locate the person that uttered a particular call sign and to identify the accompanying number and color. To model the a priori perceptual and contextual knowledge that listeners apply in a given situation, the researchers generated reference glimpses based on individual words spoken by each person. They found that for all the soundscape configurations, the person identification, location, and word recognition were much more successful than random guessing and compared favorably with how well we humans do. The duo note that real-world listeners don’t have the advantage of the reference templates. The periodicity-based glimpses nonetheless provide landmarks with high information content and could be supplemented with other, less robust auditory cues. (A. Josupeit, V. Hohmann, J. Acoust. Soc. Am. 142, 35, 2017.)
