基于谱优势与非线性变换混合的立体声声学回声消除方法

A Stereo Acoustic Echo Cancellation Method Based on the Hybrid of Spectral Dominance and Nonlinear Transformation

立体声声学回声控制系统中,两通道信号间的强相关性导致自适应算法的解不唯一,滤波器失调较大。为了解决此问题,并尽可能地保留语音质量,该文基于心理声学上的谱优势效应提出一种新型的混合立体声声学回声消除方法。根据谱优势效应,在3个低次谐频处注入能量较弱的正弦信号,以减弱低频相关性。同时对非线性变换法进行改进,以用于中高频去相关处理。与传统方法的多方面性能测试对比结果表明:所提方法能有效改善失调性能并提高收敛速度,且具有较小的语音失真。

In stereophonic Acoustic Echo Cancellation（AEC） systems, the strong correlation between the two stereophonic channels leads to nonuniqueness of adaptive solutions and further large filter misalignment. To solve this problem and preserve speech quality, the psychoacoustic spectral dominance effect is utilized to propose a novel hybrid decorrelation method for stereo AEC. According to spectral dominance, weak sinusoids are injected at the three lowest harmonics so as to reduce low-frequency coherence. Besides, the nonlinear transformation method is modified and applied to high-frequency decorrelation. Comparison test on several performances with traditional approaches is carried out. Results show that the proposed method can effectively improve filter misalignment together with convergence rate. Moreover, low speech distortion can be achieved simultaneously.