单信道语音增强中先验信噪比参数估计算法的对比分析

Comparison and Analysis of a Priori SNR Parameter Estimation Algorithm in Single Channel Speech Enhancement

先验信噪比参数估计的准确性是决定噪声背景下语音增强系统输出性能的关键因素。直接判决(Decision-Directed, DD)技术是先验信噪比估计体系中最为简易直接的算法,后续算法多为对此技术的进一步优化或改进。本文对常用的直接判决算法、两步噪声消除(Two-step Noise Reduction, TSNR)算法、改进的两步噪声消除(Modified TSNR, MTSNR)算法以及融合耦合因子(Convex Combination, CC)算法等四种先验信噪比技术进行了对比分析,给出了各种算法的基本设计原理,并从理论分析和实验仿真两个方面讨论了四种算法的输出性能及其优缺点。

The accuracy of a priori signal-to-noise ratio parameter estimation is a key factor in determining the output performance of the speech enhancement system in the noise background. The Decision- Directed (DD) technique is the simplest and straightforward algorithm in the a priori signal- to-noise ratio estimation system. Subsequent algorithms are mostly the further optimization or improvement of this technique. This paper deals with four commonly used direct decision algorithms, Two-step Noise Reduction (TSNR) algorithm, Modified Two-step Noise Cancellation (MTSNR) algorithm, and Convex Combination (CC) algorithm. A priori signal to noise ratio technology was compared and analyzed. The basic design principles of various algorithms were given. The output performance and advantages and disadvantages of the four algorithms were discussed from the aspects of theoretical analysis and experimental simulation.