双耳交互计算模型与空间听觉研究

项目名称： 双耳交互计算模型与空间听觉研究

项目编号： No.61201345

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 电子学与信息系统

项目作者： 周琳

作者单位： 东南大学

项目金额： 25万元

中文摘要： 复杂声学环境下语音信号处理的鲁棒性研究一直是重点和难点。听觉生理学和心理学研究表明人的双耳结构决定了听觉系统的鲁棒性。因此本项目基于双耳声信号处理框架，研究听觉神经对双耳声信号的交互、融合处理机制，提出相应计算模型，具体包括：研究基于子带互相关函数的空间线索，建立空间方位识别模型；建立前、后向结构的反射声抑制模型，融合均衡抵消处理过程，建立听觉系统优先效应的计算模型，实现混响环境下的空间方位识别；基于协方差矩阵实现短时子带噪声估计，在定位模型中将子带信噪比作为可信度度量，实现子带自适应选择。本项目模拟人耳听觉系统处理结构，给出完整的双耳交互计算模型，实现基于空间线索的目标语音检测和分析，提高现有语音信号处理系统对复杂声学环境的鲁棒性，为语音信号处理系统的鲁棒性研究提供新的研究框架。

中文关键词： 空间听觉；双耳空间线索；双耳声源定位；目标声源分离；

英文摘要： The robustness of speech processing system is still an important and difficult research filed. Psychoacoustic and Physiological acoustics research indicate that the binaural system of human hearing plays a major role in improving speech intelligibility. The primary goal of this project is to study the binaural perception model of human hearing and to propose the computational model of binaural interaction to improve the robustness of speech signal processing system in the noisy and reverberant environment. First, binaural cross correlation function is extracted as the localization cues for establishing sound source localization model. Second, inhibition model based on crossed fed-forward and fed-backward structure is proposed to simulate the precedence effect. Third, binaural model based on cross correlation is collaborated with equalization-cancellation model. At last, the sub-band SNR (signal-Noise Ratio) estimation based on covariance matrix is deduced to select the frequency adaptively for localization cues computation. This project will realize the target speech detection and analysis based on localization cues, and establish the framework of binaural speech signal processing. This work will improve the robustness of current speech signal processing system in real application.

英文关键词： spatial hearing；binaural spatial cues；binaural sound localization；speech separation；