中文摘要：

由于没有可比对的语音源信号，如何评价实际语音盲分离的效果非常困难。目前还没有相应的客观评价指标。将语音特征引入实际语音盲分离评价指标，提出了基于信号相关性和Mel倒谱系数高斯混合模型的听觉一独立性联合指标，客观评价了实际语音盲分离的性能。

英文摘要：

Since speech signals are unavailable, it is difficult to evaluate BSS（Blind Signal Separation） performance in the real-environment. A new separation method based on ear mechanism and independent measure is proposed. The signal correlation is incorporated into MFCC（Mel-Frequency Cepstrum Coefficient ） -based GMM（Gauss Mixed Model） to evaluate the separation performance,in which speech signals are unnecessary.