Computational Cognitive Neuroscience of Human Auditory Cortex

项目摘要 听力正常的人擅长从声音中获取有关世界的信息。我们的听觉能力 代表了令人惊叹的计算能力，直到最近才在机器中得到了任何程度的复制。 系统.然而，我们的听觉能力是非常脆弱的，在听众中受到极大的损害， 听力障碍，人工耳蜗植入和听觉神经发育障碍，特别是在 噪音的存在。承认困难往往导致挫折和社会孤立，而不是 目前的助听器、植入物和补救策略充分解决了这一问题。的长期目标 拟议的研究是揭示听觉识别的基础，并提供见解，将促进 改进的假体装置和治疗干预。开发更有效的设备， 目前，由于对现实世界认知的基础因素的不完全理解， 听力正常的听众。特别是，虽然皮层下听觉通路对声音的反应是 相对较好的研究，很少有人知道的转换发生在听觉皮层，以创造 有意义的声音结构的表征。我们建议丰富听觉识别的理解 通过三组实验来检查人类听众对真实世界声音的皮层表征， 将功能性磁共振成像（fMRI）与潜在的计算机建模相结合， 表示。目标1开发语音和音乐处理的人工神经网络模型， 将它们的表征与听觉皮层中的表征进行比较， 在模型中产生相同反应的声音的反应，并探测听觉的时间尺度， 语音和音乐分析。目标2开发和测试噪声中的音高感知模型，探索 假设音高感知受到自然声音的统计和频率的限制 耳蜗的选择性。目标3开发和测试联合定位和识别声音的模型， 利用功能磁共振成像技术研究声音的同一性和位置的大脑表征。结果将揭示 健康的听觉系统的强大的声音识别机制，并将为 对听力障碍和听觉发育障碍的皮层后果的调查， 希望能提出新的补救策略。

PROJECT SUMMARY Humans with normal hearing excel at deriving information about the world from sound. Our auditory abilities represent stunning computational feats that only recently have been replicated to any extent in machine systems. And yet our auditory abilities are highly vulnerable, being greatly compromised in listeners with hearing impairment, cochlear implants, and auditory neurodevelopmental disorders, particularly in the presence of noise. Difficulties in recognition often lead to frustration and social isolation, and are not adequately addressed by current hearing aids, implants, and remediation strategies. The long-term goal of the proposed research is to reveal the basis of auditory recognition and to provide insights that will facilitate improved prosthetic devices and therapeutic interventions. The development of more effective devices and therapies is currently limited by an incomplete understanding of the factors that underlie real-world recognition by normal-hearing listeners. In particular, although responses to sound in subcortical auditory pathways are relatively well studied, little is known about the transformations that occur within the auditory cortex to create representations of meaningful sound structure. We propose to enrich the understanding of auditory recognition with three sets of experiments that examine the cortical representation of real-world sounds in human listeners, combining functional magnetic resonance imaging (fMRI) with computational modeling of the underlying representations. Aim 1 develops artificial neural network models of speech and music processing and compares their representations to those in the auditory cortex, synthesizing and then measuring brain responses to sounds that generate the same response in a model, and probing the time scale of the auditory analysis of speech and music. Aim 2 develops and tests models of pitch perception in noise, exploring the hypothesis that pitch perception is constrained both by the statistics of natural sounds and the frequency selectivity of the cochlea. Aim 3 develops and tests models that jointly localize and recognize sounds, and probes the brain representations of sound identity and location using fMRI. The results will reveal the mechanisms underlying robust sound recognition by the healthy auditory system and will set the stage for investigations of the cortical consequences of hearing impairment and auditory developmental disorders, hopefully suggesting new strategies for remediation.