Multi-feature predictive processing of stochastic sounds in the human auditory system

项目摘要/摘要 听力正常的听者能够毫不费力地解释他们的声学环境，正在进行的解析 将声音转换成不同的来源，并通过时间追踪这些来源。为了完成这项任务，大脑 表示要与未来输入进行比较的内存中的相关感觉信息，但 这种内存表示法还不能很好地理解。通常，使用可预测来研究此机制 声音序列中的模式，听者被要求从已建立的模式中发现不同的模式。 以前的研究表明，大脑对声音的各种模式都很敏感 多个声学维度。然而，这些模式并不探索大脑如何在 自然的监听环境，在这种环境中，相关信息往往不可预测，也无法 明确而确定地表现出来。这个项目使用随机的声音序列-它展示了 统计特性而不是确定性模式--来研究大脑在多大程度上 表示在存在不确定性的情况下来自声音序列的统计信息。我们的中央 假设大脑收集高维统计信息(超出均值和方差) 捕捉跨时间和跨感知特征的不确定性，以解释正在进行的声音。在一系列中 在变化检测实验中，听者将被要求检测声音的熵的变化 沿多个感知特征变化的序列：音调、音色和空间位置。一个 将开发预测处理的计算模型来比较不同的表示法 大脑中的统计信息。模型中的感知约束将适合个人 行为，拟合的模型将被用于预测脑电异常反应 (EEG)数据。此外，感知能力的个体差异将使用单独的 任务，并将这些测量结果与模型中的结果进行比较以添加 解释权重并对模型进行改进。大脑如何处理复杂事物的计算模型 声音将打开在实验室中研究更自然、更杂乱的刺激的可能性，并且 更好地理解个体对随机声音感知的差异可能会导致更好的 用于评估时间处理能力的诊断工具。

Project Summary/Abstract Normal-hearing listeners are able to effortlessly interpret their acoustic surroundings, parsing ongoing sound into distinct sources and tracking these sources through time. To perform this task, the brain represents relevant sensory information in memory to be compared to future inputs, but the nature of this memory representation is not well understood. Typically, this mechanism is studied using predictable patterns in sound sequences, where listeners are asked to detect deviants from established patterns. Previous work has demonstrated the brain is sensitive to a wide variety of patterns in sound along multiple acoustic dimensions. These patterns, however, do not probe how the brain represents sound in natural listening environments, where relevant information is often not predictable and cannot be represented explicitly and with certainty. This project uses stochastic sound sequences—which exhibit statistical properties rather than deterministic patterns—to investigate the extent to which the brain represents statistical information from sequences of sounds in the presence of uncertainty. Our central hypothesis is that the brain collects high-dimensional statistical information (beyond mean and variance) to capture uncertainty across time and across perceptual features to interpret ongoing sound. In a series of change detection experiments, listeners will be asked to detect changes in the entropy of sound sequences varying along multiple perceptual features: pitch, timbre, and spatial location. A computational model for predictive processing will be developed to compare alternative representations of statistical information in the brain. Perceptual constraints in the model will be fit to individual behavior, and the fitted model will be used to predict deviance responses in Electroencephalography (EEG) data. Additionally, individual differences in perceptual abilities will be measured using a separate task in the same listeners, and these measures will be compared to findings from the model to add interpretative heft and improve the model. A computational model for how the brain processes complex sounds will open the possibility of investigating more natural, “messy” stimuli in the laboratory, and a better understanding of the individual differences in perception of stochastic sounds could lead to better diagnostic tools for assessing temporal processing abilities.