基于计算听觉场景分析的混叠语音分离算法  

Mixed Speech Separation Algorithm Based on Computational Auditory Scene Analysis

作　　者：王忠文[1] 谢莉萍[1] 梁杰豪[1] 

机构地区：[1]广东工业大学自动化学院,广东广州510075

出　　处：《电声技术》2008年第10期56-59,62,共5页Audio Engineering

摘　　要：提出了一种改进的混叠语音信号分离算法,该算法采用改进的黑板模型计算听觉场景分析结构,快速准确计算混叠语音信号的特征参数;利用中频特性和人的听觉感知机理,模拟了听觉神经元的发放特性,分离出不同声源的语音信号和噪声信号。通过以上2个部分的处理将输入的混叠语音信号在时频域上分割为若干有听觉意义的语音段,再聚类和重构输出不同声源的纯净语音信号。Based on the principle of CASA（Computational Auditory Scene Analysis）, the mixed speech separation arithmetic is proposed by using the model of blackboard, which can compute the parameters of mixed speech signal faster and more exact. Intermediafe frequency characteristic and auditory perceptive mechanism are used to simulate the audio nerve cell, then the mixed speech signal is separated. The speech auditory perceptive segments can be given by the processing of this part in the proposed algorithm, and they will be used for the clustering and reconstructing purified speech.

关 键 词：计算听觉场景分析 听觉感知段 语音信号分离 聚类和重构 

分 类 号：TN912[电子电信—通信与信息系统]