论文部分内容阅读
提出了一种基于χ2分布的子带噪声估计方法。带噪语音信号在临界带进行分解,并且假设子带信号服从χ2分布,然后在各个子带,采用基于χ2分布的改进最小统计量控制递归平均方法进行噪声估计。与传统的改进最小统计量控制递归平均噪声估计相比,该子带噪声估计方法可以利用人耳感知特性,并大大减少计算量。实验结果表明,提出的方法具有较好的噪声跟踪能力和较小的计算需求。采用该噪声估计的语音增强系统具有更强的噪声抑制性能和较好的增强语音信号质量。
A subband noise estimation method based on χ2 distribution is proposed. The noisy speech signal is decomposed in the critical band, and the subband signals are assumed to be subject to the χ 2 distribution. Then the noise minimization is controlled by the recursive averaging method based on χ 2 distribution in each sub-band. Compared with the traditional modified minimum-statistics control recursive average noise estimation, this sub-band noise estimation method can make use of human ear perception and greatly reduce the computational cost. Experimental results show that the proposed method has better noise tracking ability and smaller computing requirements. The speech enhancement system using this noise estimation has stronger noise suppression performance and better speech signal quality enhancement.