最小增益的语音盲分离算法中的迭代运算优化

Iterated Optimization of Minimum Gain Blind Speech Separation Algorithm

针对目前盲分离算法还无法满足对语音信号盲分离的精度需求,本文提出一种新型语音信号盲分离算法。该算法在最小增益的语音盲分离算法的基础上,针对其缺陷,引入了广义高斯分布模型,对最小增益的语音盲分离算法的迭代运算进行了优化处理,通过计算每次迭代后恢复出来的每个源的峰度值来增加原算法的分离精度。通过仿真试验进行验证,得到的结果是:改进的算法不仅仅没有失去本真效果,经过分离语音信号,对原始语音信号的波形基本保持在一定的范围内,而且表现出较好的性能。

In view of the current blind separation algorithm can not meet the accuracy requirements forblind separation of speech signals, we propose a novel algorithm for blind separation of speech signals.The algorithm is the smallest gain in speech blind separation algorithm based on, for its shortcomings,the introduction of the generalized Gaussian distribution model, the smallest gain in speech blindseparation algorithm iteration optimized process by calculating recovered after each iteration of kurtosisvalues for each source to increase the separation accuracy of the original algorithm. Simulation resultsshow that the improved algorithm, after the separation of the speech signal waveform blind maintainingthe original speech signal, virtually no distortion, good performance.