A STUDY OF VOICEPRINT RECOGNITION WITH NOISE BASED ON DUAL-CHANNEL VOICEPRINT FEATURE

Aimed at the low accuracy of voiceprint recognition under noisy environment, a dual-channel voiceprint feature (DCVF) is proposed. After the processed speech signal was passed by the Mel filter group and the Gammatone filter group, the Mel frequency cepstral coefficients (MFCC) and the Gammatone frequency cepstral coefficient (GFCC) was obtained. The mixed feature was formed by combining their dynamic differential spectrum and fusing them to form dual-channel voiceprint feature (DCVF). Experimental results show that the dual-channel voiceprint can achieve 99.5% recognition rate in pure speech data sets. Compared with MFCC and GFCC, DCVF is significantly improved on noisy speech data sets.