CRCNS: Auditory Stream Segregation and the Cocktail Party Problem

描述（由申请人提供）：听觉以一种看似毫不费力的方式进行复杂的过程和转换，统称为听觉场景分析，听觉系统通过这种方式将来自环境的声音信息整合为感知和认知经验。这里提出的项目探讨了听觉场景分析的一个基本感知组成部分，称为听觉流隔离。这种现象表现在人类和动物即使在极端嘈杂和混响的环境中也能注意到许多相互竞争的声音流之一——在文献中也被称为“鸡尾酒会问题”。尽管人类、哺乳动物、鸟类和鱼类都具有这种完全直观且无所不在的感知能力，但这种非凡的感知能力仍然笼罩在神秘之中。很少在客观的心理声学测试或在非人类物种中进行量化研究，很少在人类或动物的生理实验中进行探索。因此，开发听觉流分离计算模型的少数尝试仍然是高度推测的，并且缺乏知觉和生理数据来支持他们的公式。这反过来又极大地阻碍了工程系统中诸如自动语音识别或传感器网络中目标声音的检测和跟踪等能力的发展。

DESCRIPTION (provided by applicant): Hearing engages, in a seemingly effortless way, complex processes and transformations collectively known as auditory scene analysis, through which the auditory system consolidates acoustic information from the environment into perceptual and cognitive experiences. The project proposed here explores a fundamental perceptual component of auditory scene analysis called auditory stream segregation. This phenomenon manifests itself in the ability of humans and animals to attend to one of many competing acoustic streams even in extreme noisy and reverberant environments - also known in the literature as the "Cocktail Party Problem". While completely intuitive and omnipresent in humans, mammals, birds, and fish, this remarkable perceptual ability remains shrouded in mystery. It has been rarely quantified in objective psychoacoustical tests or investigated in non-human species, and seldom explored in physiological experiments in humans or animals. Consequently, the few attempts at developing computational models of auditory stream segregation remain highly speculative, and lack the perceptual and physiological data to support their formulations. This in turn has considerably hindered the development of such capabilities in engineering systems such as in automatic speech recognition or the detection and tracking of target sounds in sensor networks. The proposed research seeks to develop a computational model of auditory scene analysis that accounts for perceptual and neuronal-findings of auditory stream segregation. The intellectual merit of this work is providing a rigorous framework for the design of new psychoacoustic and physiological experiments of streaming, and for developing effective algorithmic implementations to tackle the "cocktail party problem" in engineering applications. The proposed research project draws upon the expertise of neurobiologists, psychoacousticians, and engineers in integrating psychoacoustic, physiological and computational techniques. The broader impact of this effort is in providing versatile and tractable models of auditory stream segregation, and hence significantly facilitating the integration of such capabilities in engineering systems, such as in automatic speech recognition or the detection and tracking of target sounds in sensor networks. This project will also provide a rigorous foundation for the design and generation of new hypotheses in order to better understand the neural basis of active listening