中文摘要：

语音信号盲分离受到随机采样的数据所影响,产生统计可靠性问题,从而影响分离的效果。Bootstrap方法是以原始数据为基础的抽样统计推断法,可方便的应用于实际的数据处理之中,分析ICA的统计可靠性,但传统Bootstrap方法本身的计算特性限制了自助样本的生成范围,从而使得自助概率分布产生了一定的偏离,使之无法渐近于真实情形。本文通过对Bootstrap样本生成范围的拓展,改进了独立分量分析在语音信号应用的可靠性算法,获得了更加精确的可靠性参数。

英文摘要：

. A major problem in the application of independent component analysis （ICA） to speech signals is that finite sample size induces statistical errors in the estimation. The statistical reliability can be investigated by Bootstrap method which is a sampling statistical interference on the base of raw data. But the conventional calculation feature of Bootstrap confined the generating range of the self-help sample, which caused some distortion on self-help probability distribution, and can not able to trene to the genuine distribution. Therefore an improving Bootstrap technique is proposed here, which can enhance the range of the Bootstrap sample and achieve the more precise reliability parameter and index of ICA estimation on speech signals.