作者:兰朝凤，陈英淇，林小佳，刘岩，陈旭奇
\n

Authors:LAN Chao-feng,CHEN Ying-qi,LIN Xiao-jia,LIU Yan,CHEN Xu-qi
\n
摘要:摘要:随着语音处理技术的发展，新的语音分离算法不断地被提出。针对观测信号中噪声而导致分离效果不理想的问题，将几何运算(GA)方法和快速独立成分分析(FastICA)算法结合提出了GA_FastICA算法。为探究GA_FastICA算法的有效性，绘制了分离后语音信号的时域波形，给出了分离前后语音信号的相关系数。当信噪比为4dB时，分离后语音信号与原始语音信号的相关系数为0.7852。仿真实验结果表明，在信噪比为12 dB，factory、babble噪声条件下，GA_FastICA算法相较于FastICA算法相关系数提高了0.0212和0.0304；信噪比为8dB的条件下，相关系数提高了0.1374和0.1328。GA_FastICA算法可有效分离语音信号，在噪声环境下具有较好的语音分离效果。
Abstract:Abstract:With the development of speech processing technology, new speech separation algorithms are constantly proposed. The GA_FastICA algorithm is proposed by combining the Geometric Approach (GA) algorithm and Fast Independent Component Analysis (FastICA) algorithm for the problem of unsatisfactory separation due to the noise in the observed signal and combining the geometric operation method. The time domain waveforms of the separated speech signals are plotted, and the correlation coefficients of the original and separated speech signals are given to investigate the effectiveness of the GA algorithm.When the signal-to-noise ratio is 4dB, the correlation coefficient of the separated speech signal and the original speech signal is 0.7852 . The experimental simulation results show that under the signal-to-noise ratio of 12dB, factory and babble noise conditions, the GA_FastICA algorithm improves the correlation coefficient by 0.0212 and 0.0304 compared with the FastICA algorithm, and the correlation coefficients were improved by 0.1374 and 0.1328 for a signal-to-noise ratio of 8dB. The GA_FastICA algorithm can effectively separate the speech signal, and the noisy environment GA_FastICA algorithm can effectively separate speech signals and has a better speech separation effect.

PDF全文下载地址:

可免费Download/下载PDF全文