基于CTF-GSC和后置滤波的麦克风阵列语音增强

发布时间：2018-08-27 13:49

【摘要】：由于语音环境中各种噪声的特殊性、复杂性和不确定性等因素,尤其在带有混响的复杂环境下,一般的广义旁瓣抵消器语音增强算法的性能严重下降,很难取得较为理想的抑制效果。针对这一情况,该文提出了一种基于改进型归一化最小均方的卷积传递函数广义旁瓣抵消器和改进型后置滤波语音增强算法。该算法利用卷积传递函数近似形式代替传递函数广义旁瓣抵消器中的相乘传递函数近似形式,并结合后置维纳滤波进行语音增强。实验测试结果表明,该算法可以有效抑制相干性和非相干性噪声,提高系统信噪比,减少语音失真。在含有混响的复杂环境中,与其他经典算法对比,该算法的鲁棒性更好,效果更明显。
[Abstract]:Because of the particularity, complexity and uncertainty of various noises in speech environment, especially in the complex environment with reverberation, the performance of the general generalized sidelobe canceller speech enhancement algorithm is seriously reduced. It is difficult to obtain more ideal inhibition effect. In this paper, an improved generalized sidelobe canceller for convolution transfer function and an improved speech enhancement algorithm based on improved normalized least mean square are proposed. In this algorithm, convolution transfer function approximation is used to replace the approximate form of multiplicative transfer function in transfer function generalized sidelobe canceller, and speech enhancement is carried out with post-Wiener filter. The experimental results show that the proposed algorithm can effectively suppress coherent and non-coherent noises, improve the signal-to-noise ratio of the system and reduce the speech distortion. In the complex environment with reverberation, compared with other classical algorithms, the algorithm is more robust and more effective.
【作者单位】：湖南大学电气与信息工程学院;
【基金】：湖南省科技计划重点基金资助项目(2015JC3053) 湖南省自然科学基金资助项目(2017JJ2052) 教育部产学合作协同育人基金资助项目(201601004010) 湖南省普通高校教学改革研究基金资助项目(湘教通[2016]400号)
【分类号】：TN912.35

【相似文献】