基于增益函数的线性预测语音增强算法的研究

增益函数估计的精确度直接决定了语音增强算法的性能,并且增益函数的估计依赖于先验信噪比估计算法.DD(Decision Directed)算法具有较低的计算复杂度和有效抑制音乐噪声的优点,是最常用的先验信噪比估计算法,但是DD算法具有语音失真严重的缺点.介绍了DD算法、MDD(Modified method)算法和线性预测增益函数算法,理论分析了各自性能的优缺点,并进行了实验验证.实验结果显示,增益函数的线性预测方法可以取得较好的语音增强性能.

Research of Speech Enhancement AlgorithmsBased on linear prediction of gain function

The accuracy of gain function estimation directly determines the performance of speech enhancement algorithm, and gain function estimation depends on the a prior SNR estimation method. DD (Decision Directed) method has the advantages of lower computational complexity and effectively restrain music noise, which is the mostly commonly used a prior SNR estimation approach, however DD algorithm has a serious shortcoming of speech distortion. In this paper, DD method, MDD (Modified method) algorithm and linear prediction gain function approach introduced, in the same time, we made theory analyzes the advantages and disadvantages of their performance, and experimental verification. The experimental results show that the linear prediction of gain function approach has achieved better performance of speech enhancement.