Midbrain Computational and Robotic Auditory Model for focused hearing (MiCRAM)

我们的目标是在计算机模型中复制听觉中脑（下丘）的处理过程，以分析声音。我们的假设是，基于大脑处理声音的方式，一个受生物学启发的前皮层处理模型将比现有的计算模型更有效。我们的听力上级基于计算机的声音识别系统的一个方面是它能够在嘈杂的环境中分离和识别声音;例如，当我们在嘈杂的聚会上进行对话时。如果我们能够通过计算机接口或机器人系统实现这种水平的性能，那么生活质量将有很大的好处。特别是，我们打算证明我们的模型的实用性，通过调整它来控制机器人，能够在嘈杂的环境中响应声音刺激。反过来，我们希望我们的模型能够产生关于大脑功能的预测。我们的目标是通过计算机、机器人、听觉处理和脑科学专家团队的合作来实现这一目标。当声音激活我们耳朵中的受体时，我们就能听到周围世界的声音。这些感受器将声音编码成电脉冲，激活大脑中的一系列处理中心，最终激活皮层。耳蜗是耳朵中感知声音的部分，它的形状像蜗牛壳。沿着它的长度是一排内毛细胞。每个细胞对特定的频率最敏感，这些毛细胞反过来激活听觉神经纤维。由于每个神经纤维只对一个狭窄的声音频率范围做出反应，并以有序的方式投射到大脑，因此频率以空间有序或地形的方式表示，称为音调位置表示。神经纤维携带的信息进入大脑，并分成许多处理通道，强调声音流的不同方面。这些单独的音调表征在下丘（IC）集中并处理，下丘又将输出发送到丘脑，然后发送到听觉皮层。有证据表明，由于这些声音流在IC处汇聚，因此在此水平上有足够的信息可以识别声音是什么以及它来自哪里。IC还接收来自皮层的反馈。这可以提供期望的声音信号模式，下丘将该声音信号模式与传入的声音模式进行比较以识别差异。为了建立我们的模型，我们团队的神经科学家将构建一个关于听觉脑干的布线和连接的现有知识的数据库，并在必要时添加关于动物脑干听觉处理的新实验。建模人员将使用这些信息来指导创建可以控制机器人动作的计算机模型。我们将使用机器人，因为动物不只是被动地听，而是主动地寻找声音，以建立他们的世界的听觉图像。这种积极的行为使他们能够根据声音的处理方式将具有不同结果的实验场景放在一起。与机器人一起工作使得研究结果更有可能在帮助建立更好的助听器、语音理解系统、声音跟踪系统和声音控制机器人方面有实际用途。我们希望我们的模型能帮助我们预测中脑的功能，我们将在动物和机器人实验中测试这些预测。声音处理建模的专家们非常有兴趣测试一些关于耳朵和大脑如何处理声音的理论是否与生物学中看到的一致，因此我们的建模将有助于指导未来研究的方向并减少动物使用。我们开发的数据库将提供给该领域的其他研究人员。

Our aim is to replicate in a computer model the processing that occurs in the auditory midbrain, the inferior colliculus, to analyse sounds. Our hypothesis is that a biologically inspired model of precortical processing, based on the way the brain processes sounds, will be more efficient than existing computational models. One way in which our hearing is superior to computer based sound recognition systems is in its ability to separate and identify sounds in noisy environments; for example when we hold a conversation at a noisy party. There would be great benefits to the quality of life if we could achieve this level of performance with computer interfaces or robotic systems. In particular we intend to demonstrate the utility of our model by adapting it to control a robot that is able to respond to sound stimuli in a noisy environment. In turn, we expect our model to generate predictions about brain function. We aim to do this through a collaboration between a team experts in computing, robotics, auditory processing and brain science.We hear sounds in the world around us when they activate receptors in our ears. These receptors encode sound to electrical impulses that activate a chain of processing centres in the brain and eventually the cortex. The cochlea, the part of the ear where sound is sensed, is shaped like a snail shell. Running down its length is a row of inner hair cells. Each cell is most sensitive to a particular frequency, and these hair cells in turn activate auditory nerve fibres. Because each nerve fibre only responds to a narrow range of sound frequencies and projects in an orderly way to the brain, frequency is represented in a spatially ordered or topographic manner called a tonotopic representation.The information carried by the nerve fibres enters the brain and divides into a number of processing channels that emphasise different aspects of the sound stream. These individual tonotopic representations converge and are processed in the inferior colliculus (IC), which in turn sends outputs to the thalamus and then to the auditory cortex. Evidence suggests that because these streams converge at the IC there is sufficient information available at this level to identify what the sound is and where it comes from. The IC also receives feedback from the cortex. This may provide an expected pattern of sound signals that the inferior colliculus compares with the incoming sound pattern to identify differences. It also may be that this acts in some way to spotlight auditory attention to emphasize some sounds and de-emphasise others.To build our model, the neuroscientists in our team will construct a database of current knowledge about the wiring and connections of the auditory brainstem, and add to this where necessary with new experiments about auditory processing in the brainstem of animals. The modellers will use this information to guide the creation of a computer model that can control the actions of a robot. We will use robots because animals do not just passively listen but actively seek out sounds to build an auditory picture of their world. This active behaviour allows them to put together experimental scenarios that have different results depending on how sound is processed. Working with robots makes it more likely that the results will have practical uses in helping to build better hearing aids, speech understanding systems, sound tracking systems, and sound-controlled robots. We expect that our model will help us to make predictions about how the midbrain functions and we will test these predictions in animal and robot experiments. The experts in modelling sound processing are very interested in testing whether some theories on how the ear and brain handle sound are consistent with what is seen in biology and thus our modelling will help to guide the direction of future research and reduce animal use. The database we develop will be made available to other researchers in the field.