【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。更多还原

【Abstract】 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.Intermediate 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.更多还原