一种融合相位的先验信噪比算法

A Priori Signal-to-Noise Ratio Algorithm Based on Phase

研究表明,大多数语音增强算法都存在一个核心参数—先验信噪比,常用的估计先验信噪比的算法是直接判决法(Decision-Directed,DD)。然而,传统的DD算法是在纯净语音谱与噪声谱相互正交的前提条件下完成,其结果会导致增强后的语音质量受损。为了更加准确地估计先验信噪比,利用三角函数关系,提出一种融合相位影响的先验信噪比估计。仿真实验显示,所提的算法能够明显提升语音增强质量。

Research shows that for most speech enhancement algorithms,there is a core parameter-a priori signal to noise ratio.The commonly used algorithm to estimate the priori signal to noise ratio is Decision-Directed(DD)algorithm.However,the traditional DD algorithm is completed on the premise that the pure speech spectrum and the noise spectrum are orthogonal to each other,and the result will cause the enhanced speech quality to be damaged.In order to estimate a priori Signal Noise Ratio(SNR)more accurately,a priori SNR estimation based on phase effect is proposed by using trigonometric function.Simulation results show that the proposed algorithm can significantly improve the quality of speech enhancement.