一种基于模糊逻辑推理的双端发声检测算法

Doubletalk Detection Algorithm Based on Fuzzy Inference

为了提高双端发声检测算法的性能、减少切音现象,文中提出了一种基于双统计量模糊逻辑推理的双端发声检测算法,并在回声消除系统中加入了舒适噪声生成部分。文中对传统的双端发声检测算法进行了分析,指出了这些算法由于采用固定门限,不能处理双端发声检测统计量其固有的模糊性,造成了在给定的虚警概率下,漏警概率偏高的问题。算法分析和实验结果表明,该算法大大地提高了双端发声检测的性能,很好地解决了加入非线性处理后的切音问题。

The doubletalk detection in echo cancellation system detection is presented. The traditional doubletalk detection algorithms are analyzed. It is pointed out that because these algorithms usually adopt fixed thresholds, they can hardly process the large variances of detection statistics, so that a high probability of miss-detection is introduced. If a nonlinear process for residual signal is conducted, several near-end speeches can be falsely cut. To tackle this issue, a novel double-talk detection algorithm with two detection statistics based on fuzzy inference is proposed. Furthermore, the comfort noise generation is integrated into the echo cancellation scheme as well. Experimental results show that the proposed algorithm improves the performance of double-talk detection, and the problem of near-end speech cut is solved.