摘要
对SIMO系统来说,盲语音去混响过程就是将接收到的未知多通道输出的卷积信号进行盲反卷积并得到源信号的过程。盲反卷积需要先盲辨识多信道脉冲响应,再对信号反卷积。在时域上,可以利用卡尔曼滤波器(KF)同步对SIMO系统及其逆系统进行盲多信道识别来实现盲反卷积,但是计算量较大。在深入分析时域算法的基础上,将部分时域计算变换到频域计算,通过快速傅里叶变换(FFT)代替卷积计算,从而大大降低算法的总计算量。仿真实验表明,用部分频域算法取代时域算法,在保证算法稳定性的同时可以明显改善算法的收敛速度,降低算法的复杂度。
For SIMO system,the blind speech dereverberation process is the process of blindly deconvolving the received convo⁃lution signal of unknown multi-channel output to obtain the source signal.Blind deconvolution requires the blind identification of the multi-channel system firstly.In time domain,blind deconvolution can be realized by the blind multi-channel identification of the SI⁃MO system and its inverse system synchronously by using Kalman filter(KF),but the amount of computation is large.On the basis of in-depth analysis,the time-domain algorithm is transformed partly into a frequency-domain algorithm,i.e.,those convolution calculations are replaced with the Fast Fourier Transform(FFT),thereby the total calculation amount of the algorithm is reduced greatly.Simulation experiments show that replacing the time-domain algorithm with some frequency-domain algorithms can significant⁃ly improve the algorithm's convergence speed and reduce the algorithm's complexity while ensuring the stability of the algorithm.
作者
董吉汉
梅铁民
DONG Jihan;MEI Tiemin(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang 110159,China)
出处
《通信与信息技术》
2022年第5期75-79,共5页
Communication & Information Technology