Neural underpinnings of bottom-up and top-down auditory attention in real-life environments

我们有能力一次专注于一项听力任务，并将周围的其他声音置于背景中。这种自发的、自上而下的注意力指向最相关的声音流，可能会被显著的背景声音事件（如汽车喇叭）打断--这种自下而上的机制使我们能够及时对周围的事件做出反应。根据周围的听觉场景和听者的认知状态，一些显著事件可能未被检测到。这种情况经常发生在我们的日常生活中，因为我们倾向于忽略与我们无关的声音。然而，有时候，没有检测到一个显著的声音可能会导致严重的后果。理解这种常见的神经过程的基础，如背景显著事件的检测，本身就非常重要。此外，只有通过理解这些过程，才有可能提出替代警告的解决方案，如果相关的显着事件没有被监听器检测到。大多数关于听觉注意的发现来自于高度受控的实验室实验。本研究的目的有两个：（1）研究真实的生活中显著事件的神经反应及其对主要听力任务的影响;（2）探索在脑-机接口（BCI）中对检测到的和未检测到的显著事件进行近真实的时分类的可能性。我将在三个独立的脑电图（EEG）研究中实现这些目标，其中人类参与者将听取背景中出现的突出声音的演讲。由于我的目标是研究真实的生活条件下的神经过程，因此所有三项研究都将使用不显眼的移动的EEG设置进行。我将在受控的实验室条件下使用自然主义刺激开始我的研究，获得与以前文献中的结果相当的结果。我的下一步将是研究实验室外自然发生的、不受控制的显著事件的神经反应，这是由于最近开发的显著性模型可以自动估计听觉场景中事件的显著性。一旦已知显著事件检测的神经相关性，第三个研究将探讨听觉脑机接口对未检测到的显著事件提供反馈的可行性，我将采用两种互补的分析方法--一种新奇的P3分析方法来研究背景显著事件的神经反应，另一种新颖的听觉包络注意跟踪方法来观察显著事件对顶部的影响，向下注意前景音频流。将这两种方法应用于EEG数据可能会导致显著事件检测的鲁棒神经相关性，从而为日常场景提供设计可靠的闭环听觉BCI的方法。

We have an ability to focus on one listening task at time and push other surrounding sounds into background. This voluntary, top-down attention directed towards the most relevant sound stream can be interrupted by salient background sound events such as car horn - a bottom-up mechanism that allows us to timely react to events from our surroundings. Depending on the surrounding auditory scene and cognitive state of a listener some salient events can go undetected. This happens often in our daily routine, as we tend to ignore sounds irrelevant to us. Sometimes, however, not detecting a salient sound may be followed by severe consequences. Understanding the basis of such a common neural process as detection of a background salient event is on its own of great importance. Furthermore, only by understanding these processes it is possible to come up with solutions for alternative warnings if a relevant salient event is not detected by a listener.Not much is currently known about salient event processing in natural situations. Most findings on auditory attention come from highly-controlled laboratory experiments. The goal of this project is two-fold: (1) to investigate neural responses to salient events in real life and their influence on main listening task and (2) to explore the possibility of near-real-time classification of detected vs. undetected salient events for application in brain-computer interfaces (BCIs). I will achieve these goals in three separate electroencephalography (EEG) studies where human participants will be listening to speech with salient sounds occurring in background. As my aim is to study neural processes in real life conditions, all three studies will be conducted using an unobtrusive mobile EEG setup. I will start my investigation in controlled laboratory conditions using naturalistic stimuli, obtaining results comparable to those in previous literature. My next step will be to investigate neural responses to naturally occurring, uncontrolled salient events outside of laboratory, which is made possible by recent development of saliency models that can automatically estimate salience of events in an auditory scene. Once the neural correlates of salient event detection are known, the third study will investigate feasibility of an auditory BCI that could provide feedback on undetected salient events.I will apply two complementary analysis approaches- a novelty P3 analysis to investigate neural responses to background salient events and novel auditory envelope attention tracking approach to observe the influence of salient events on the top-down attention to foreground audio stream. Applying both methods to EEG data is likely to lead to robust neural correlates of salient event detection to the point of providing means of designing a reliable closed-loop auditory BCI for every-day scenarios.