Using and ignoring acoustic feature differences in auditory object recognition.

摘要

Object recognition is crucially important to understanding complex auditory scenes. We do it effortlessly, yet it is dependent not only on extracting a large number of acoustic features, but knowing which ones to use when. In certain settings, auditory objects must be recognized despite variations in some features, a phenomenon known as perceptual invariance. In other settings, objects are selected from competing sound sources by exploiting separations in acoustic features. For this dissertation, we used a combination of behavioral experiments and neural recordings to measure the effects of altering acoustic features on object recognition. In the first set of experiments, we time-stretched and compressed natural stimuli and presented them to zebra finches. Behavioral results with trained birds showed that song identity could be determined despite these perturbations. Electrophysiological recordings in zebra finch field L (the avian homolog of mammalian primary auditory cortex) revealed that, while time-warp invariance does not exist per se, responses to time-warped stimuli still contained more than sufficient information to identify target songs. In the next set of experiments, we presented human listeners with competing spoken digits that had specific locations and pitches. We instructed listeners to use one feature and ignore the other to report the digits that matched a primer phrase. Performance improved when the separation of the target and masker in the task-relevant feature increased, but the continuity of the task-irrelevant feature also influenced performance in some cases. These results indicate that task-relevant and task-irrelevant features are perceptually bound together, consistent with the idea that auditory attention operates on objects. In the final set of experiments, we investigated how spatial location affects responses in field L. We found that the location of birdsong presented in quiet had little effect on responses; however, when we presented the songs with a masker, spatial effects were dramatic, with neural performance much higher at some spatial configurations than at others. The locations of both the target and masker sounds were important, and spatial response patterns were diverse among neurons.