Temporal Pattern Perception Mechanisms for Acoustic Communication

项目概要/摘要： 处理声学通信信号是最困难但最重要的能力之一， 听觉系统这些能力是语言和语音处理的核心， 成功或失败对整个生命周期的生活质量有着深远的影响。了解 支持这些基本能力的神经生物学机制有望促进 辅助听力设备，以及改善对学习障碍的诊断和治疗 和沟通障碍，如听觉处理障碍，诵读困难，和特定的 语言障碍人类的非侵入性神经科学技术揭示了 语言相关的处理，但不回答如何个别神经元和神经回路 实现语言相关计算。因此，电路级神经计算机制 支持声学通信信号处理的技术仍然知之甚少。多行 的研究表明，鸣禽可以提供一个很好的模型， 与语言相关的听觉处理能力。这项提案研究神经机制 鸣禽和人类的听觉时间模式处理能力。在Aim中 1，我们测试了一个强大的建模框架的细胞级预测，称为预测 编码，提出作为一种通用的计算机制，以支持学习识别 在多个时间尺度上的复杂的时间模式化信号。我们将最先进的联合收割机 机器学习方法与多电极电生理学，以测试明确的模型， 单个皮层神经元和神经元中的自然刺激表示、预测和错误编码 人口。听觉感知的一个方面是语音的离散化 转化为习得的分类感知声音（音素）。在目标2中，我们使用预测编码 框架，以调查自然听觉类别的学习分类感知， 皮层神经元的数量。在人类中，相邻音素之间的转换统计 可以帮助或改变音素分类，为语言学习者和听众提供线索， 消除感知上相似的声音。Aim 2还研究了分类神经元 表示受时间上下文的影响。除了哪些音素出现在一个 序列，语音处理还需要知道这些元素出现的位置。的敏感性 语音序列的统计学特征早在婴儿学会说话之前就已经建立， 并在整个成年期继续影响认知和理解。鸣禽也 注意他们声音交流信号中的统计学意义。在目标3中，我们专注于 序列特异性信息是如何被单个神经元和神经群体编码的， 听觉皮层拟议的办法可在近期内取得进展， 基础语言相关能力的基本神经生物学基础和一般的 在这个框架内，可以提出更复杂的，独特的人类过程， 最终经过测试。

Project Summary/Abstract: Processing acoustic communication signals is among the most difficult yet vital abilities of the auditory system. These abilities lie at the heart of language and speech processing, and their success or failure has profound impacts on quality of life across the lifespan. Understanding the neurobiological mechanisms that support these basic abilities holds promise for advancing assistive listening devices, and for improving diagnoses and treatments for learning disabilities and communication disorders, such as auditory processing disorder, dyslexia, and specific language impairment. Non-invasive neuroscience techniques in humans reveal the loci of language-related processing but do not answer how individual neurons and neural circuits implement language-relevant computations. Thus, circuit-level neuro-computational mechanisms that support acoustic communication signal processing remain poorly understood. Multiple lines of research suggest that songbirds can provide an excellent model for investigating shared auditory processing abilities relevant to language. This proposal investigates neural mechanisms of auditory temporal pattern processing abilities shared between songbirds and humans. In Aim 1, we test the cellular-level predictions of a powerful modelling framework, called predictive coding, proposed as a general computational mechanism to support the learned recognition of complex temporally patterned signals at multiple timescales. We combine state-of-the-art machine learning methods with multi-electrode electrophysiology, to test explicit models for natural stimulus representation, prediction, and error coding in single cortical neurons and neural populations. One aspect of auditory perception integral to speech is the discretization of the signal into learned categorically perceived sounds (phonemes). In Aim 2, we use the predictive coding framework to investigate the learned categorical perception of natural auditory categories in populations of cortical neurons. In humans, the transition statistics between adjacent phonemes can aid or alter phoneme categorization, providing cues for language learners and listeners to disambiguate perceptually similar sounds. Aim2 also examines how categorical neural representations are affected by temporal context. In addition to which phonemes occur in a sequence, speech processing also requires knowing where those elements occur. Sensitivities to the statistical regularities of speech sequences are established long before infants learn to speak, and continue to affect both recognition and comprehension throughout adulthood. Songbirds also attend to the statistical regularities in their vocal communication signals. In Aim 3, we focus on how sequence-specific information is encoded by single neurons and neural populations in auditory cortex. The proposed approach permits progress in the near term towards establishing the basic neurobiological substrates of foundational language-relevant abilities and a general framework within which more complex, uniquely human processes, can be proposed and eventually tested.