De-Noising Using Dual Threshold Function For Speaker Recognition At Low SNR

In recent years, speaker recognition as biometric authentication has been attracting attention. In automatic speaker recognition system, there are some key steps such as de-noising, feature extraction, creating speaker models and machine learning. Feature extraction is one of the most important processes for improving the accuracy of the speaker recognition. However, the performance of feature extraction is reduced by impact of noise. Especially, the noise influence is large at low SNR. In this paper, we propose a new method for de-noising method using a dual threshold reduction function. At low SNR environment, the proposed method has good performances compare with other threshold function methods.