融合耳蜗—神经元响应机制的听觉建模及语音识别应用研究

Modeling auditory perception and developing corresponding information processing methods are frontier topics in the fields of cognitive and neuroscience, information and artificial intelligence. At present, there is a lack of a method to fuse the flow of auditory information with the information processing mechanism so that the information of neuron filter response can reflect the characteristics of auditory information processing. This project focuses on the bio-neuron filtering mechanism, and classifies the biological processing of environmental sound information into the process of "cochlear perception-neuron filtering". It simulates the auditory information processing process from the macro-information flow and micro-processing mechanism. Specific contents include: focusing on the transmission form of information in cochlea and neurons, using vibration method to establish cochlear perception model and neuron filtering model; researching the response, conversion and maintenance characteristics and mechanism of auditory nerve information; putting forward a new auditory information processing method with good consistency of auditory information response and simple realization form; and designing psychoacoustics. Psychoacoustics and simulation experiments are designed to verify the consistency between speech recognition and cochlear auditory information processing in multi-sound perception environment. The research of this project will open up a new field of science and application of auditory information processing, help to promote the cross-integration of information, control, artificial intelligence, neuroscience and so on. It is of great significance to the research and development of high-tech such as the design of acoustic sensors.

对听觉感知建模并发展相应的信息处理方法是认知与神经科学、信息和人工智能领域的前沿性课题。当前缺乏一种将听觉信息流向与信息处理机制相融合使神经元滤波响应的信息能够反映听觉信息处理特性的方法。本项目着眼于生物神经元滤波机制，将生物对外环境声音信息进行处理归纳为“耳蜗感知—神经元滤波”过程，从宏观信息流向和微观处理机制完整模拟听觉信息处理过程。具体内容包括：着眼于信息在耳蜗与神经元内的传递形式，采用振动学方法建立耳蜗感知模型、神经元滤波模型；研究听神经信息的响应、转换、保持特性和机理；提出一种具有良好听觉信息响应一致性且兼具简单实现形式的听觉信息处理新方法；并设计心理声学和仿真实验，验证该模型和方法在多声感知环境下语音识别与耳蜗处理听觉信息的一致程度。该项目研究将开辟听觉信息相关全新的科学和应用领域，有助于促进信息、控制、神经科学等的交叉融合，对人工智能、声传感器等高新技术的研发具有重要意义。

本项目基本完成了原计划的研究目标。初步建立了耳蜗感知模型三维架构，在对听觉机理的生物物理学分析基础上，提出了基于等效振动系统分析研究耳蜗基底膜振动以及耳蜗内微观力学特性的分析方法；提出了通过将生物神经元滤波机制引入振动理论建立基于振动理论的神经元滤波模型，分析并揭示了听觉信息处理过程的频率位置分布特性、高听敏度特性以及主动非线性放大机制；设计了一系列心理声学和仿真实验，检验了耳蜗感知—神经元滤波模型在复杂多声感知环境下的语音识别与耳蜗处理听觉信息的一致程度，获得了一系列包括语音频谱、语音复现时间等相关数据，建立了中文语音数据库，为获得增强语音信号的主观听觉感受和客观评价指标的建立提供了关键基础。项目执行期间，课题组完成研究论文6篇，其中1篇发表在Journal of Biology System (SCI收入期刊)；1篇被Frontiers In Psychology (SSCI收入期刊) 接受发表；1篇论文投稿在Physical Review Physics Education Research (SCI和SSCI收入期刊，教育学类一流期刊)，目前按审稿意见修改中；另有3篇论文即将完成投稿。依托本项目，招收和培养了研究生3名。