用RBF网解决说话人确认中的信道非线性失真

Using RBF Network to Solve the Channel Nonlinear Distortion in Speaker Verification

提出了一种基于RBF神经网络的非线性失真补偿技术。该方法利用RBF网络的非线性映射特性,在特征域对传输信道中的非线性失真进行补偿,较好地解决了现有失真补偿技术对非线性失真补偿效果不佳的缺点,降低了信道非线性失真对说话人确认系统的影响。实验结果显示,不论在有线性失真或是非线性失真时该方法都能较好地补偿失真导致的差异,保持说话人确认效果。

A novel approach of nonlinear distortion compensation technique based on RBF network is proposed here. Utilizing the nonlinear mapping characters of RBF,this approach can compensate the nonlinear distortion in transmission channel to minimum the mismatch between training and testing.The speaker verification experimental results show that performance decline caused by channel distortion can be approximate eliminated.