Application of muti-channal subspace algorithm to speaker recognition
Qingning Zeng · Shengxue jishu · 2008
The multi-channel signal subspace algorithm for speech enhancement is studied, and the algor- ithm's parameters which affects the performance are analyzed.A simple and an effective method for esti- mating the noise variance are given. Through analysis, it is proved that the multi-channel signal subspace algorithm can denoise obviously and make very little voice injury. Compared with the single-channel subspace speech enhancement algorithm,the multi-channel signal subspace algorithm can not only upgrade the per- formance, but also have no increase in the amount of calculation. The muti-channel signal subspace alg- orithm has been used in speaker recognition system, and compared with other speech enhancement alg- orithms, such as Delay and Sum Beamfonning, Beamforming and Wiener Filter, and Linear Constraint Minimum Variance Beamfonning.The experimental results show that the muti-chanel signal subspace alg- orithra can effectively improve the recognition rate of the speaker recognition system in noisy environment.