The Impact of Low-Pass Filter in Speaker Identification

Speaker identification model commonly uses Mel Frequency Ceptral Coefficient (MFCC) and Gaussian Mixture Model (GMM). Due to many weaknesses from previous studies for noised speech, here a low-pass-filter is proposed to reduce the high-frequency signals. The low-pass-filter is expected to calculate cut-offs. Experimental results show that the low-pass filter significantly improves the accuracy of sound detection for the high noised signal.