Auditory Scene Analysis and Temporal Cortical Computations

 描述(申请人提供)：当一个房间里的许多人同时说话时，他们的声音在到达我们的耳朵之前就相互混合了。尽管把声音混合成单独的声音是一个极其困难的数学问题，但我们的大脑通常会完成这项任务，而且通常不需要付出多少明显的努力。这项艰巨任务的神经基础根本没有得到很好的理解。 此外，当这种能力下降时，例如由于听力损失或衰老，不知道神经处理的哪些特定机制在保留这种能力的剩余方面是最关键的。为了解决这些问题，本研究建议使用脑磁图(MEG)从行为受试者的听觉皮层记录，特别是对单个声音元素及其混合物的时间动态神经反应。将神经反应与它们的听觉刺激和注意力状态联系起来，我们就可以推断出这些声音的神经表征。这些神经表示是暂时的：神经处理随着正在进行的声学动力学而及时展开。这项研究计划将使用这些时间表征来研究复杂的听觉场景是如何进行神经编码的，从整个声学场景的广泛混合到独立的单个来源，在听觉皮质的不同区域，并特别强调语音。其最重要的假设是，听觉皮质使用一种真正的时间性质的通用神经编码方案，这不仅是一般听觉处理的基础，也是听觉场景分离的基础。第一个具体目标将决定听觉皮质如何在困难的听力环境中神经地表达语音。一个例子是混响环境中的噪声中的语音，这是一个非常相关的组合，可以强烈地破坏语音的清晰度。另一个例子是在几个相互竞争的发言者在场的情况下听取发言者的意见。在这种情况下，了解背景(相互竞争的说话者的混合)是如何在神经上表示的是特别感兴趣的，并且对于确定大脑如何将前景语音从背景中分离出来是直接相关的。第二个具体目标将确定动态非语音声音的这些神经语音表示的类比，特别是当声音是较大声学场景的单独组件时。这将把有关语音分离的已知内容推广到更广泛的声音类别(虽然语音对人类听众非常重要，但大多数声音都不是语音)。第三个具体目标是研究听觉皮质在复杂的声学场景中识别和隔离单个说话人的详细神经机制。基音和音色这两个已知对这项任务很重要的声学线索被单独和独立地修改，以便可以确定它们对语音听觉场景分离的神经过程的单独贡献。

 DESCRIPTION (provided by applicant): When many people in a room are talking at the same time, the sounds of their voices mix with each other before ever arriving at our ears. Despite the fact that sorting out this sound mixture into individual voices is a profoundly difficult mathematical problem, our brain routinely accomplishes this task, and often with little apparent effort. The neural underpinnings of this nonetheless difficult task are not at all well understood. Furthermore, when this ability declines, e.g., due to hearing loss or aging, it is not known which specific mechanisms of the neural processing are the most critical in preserving remaining aspects of this ability. In order to address these issues, this proposed research uses magnetoencephalography (MEG) to record from the auditory cortex of behaving human subjects, specifically the temporally dynamic neural responses to individual sound elements and their mixtures. Linking the neural responses with their auditory stimuli and attentional state allows us to infer neural representations of these sounds. These neural representations are temporal: the neural processing unfolds in time in response to the ongoing acoustic dynamics. This research program will use these temporal representations to investigate how complex auditory scenes are neurally encoded, from the broad mixture of the entire acoustic scene to separated individual sources, in different areas of auditory cortex, and with a special emphasis on speech. Its overarching hypothesis is that auditory cortex employs a universal neural encoding scheme, genuinely temporal in nature, which underlies not only general auditory processing but also auditory scene segregation. The first specific aim will determine how auditory cortex neurally represents speech in difficult listening situations. One example is of speech in noise in a reverberant environment, a very relevant combination which can strongly undermine speech intelligibility. Another example is listening to a speaker in the presence of several competing speakers. In this case, understanding how the background (the mixture of the competing speakers) is neurally represented is of particular interest, and of direct relevance in determining how the brain segregates the foreground speech from the background. The second specific aim will determine analogs of these neural speech representations for dynamic non-speech sounds, especially when the sounds are separate components of a larger acoustic scene. This will generalize what is known about speech segregation to a wider class of sounds (while speech is very important for human listeners, most sounds are not speech). The third specific aim investigates the detailed neural mechanisms by which auditory cortex identifies and isolates individual speakers in a complex acoustic scene. Pitch and timbre, two acoustic cues known to be important for this task, are separately and independently modified, so that their individual contributions to the neural process of auditory scene segregation of speech may be determined.