自适应零限波束形成语音增强算法鲁棒性分析

Robustness analysis of adaptive null-forming scheme for speech enhancement

分析了双传声器自适应零限波束形成语音增强算法对传声器不一致和本底噪声的鲁棒性。结果表明:信干比越高,算法对传声器不一致的鲁棒性越差;当信干比很低时,算法对相位不一致是鲁棒的。幅度不一致会降低算法对干扰的抑制能力,但引起的目标信号失真很小。相对于传声器不一致,本底噪声对算法性能的影响较小。对自适应滤波器权值加约束和对传声器做校准可以提高算法性能。

The robustness of adaptive null-forming scheme with two microphones for speech enhancement was evaluated. The analysis reveals that the algorithm is sensitive to errors in the assumed signal model when the Signal-to-InterferenceRatio is high. However, when the Signal-to-Interference Ratio is low, the algorithm is robust to the phase mismatch between the two microphones. The magnitude mismatch decreases the amount of interference reduction, while causes little target distortion. Compared with the mismatch, the noise level of microphones affects the performance of the algorithm less. Constrain of filter weight or calibration of the microphones will improve the robustness of the adaptive null-forming algorithm.