Abstract—To address the problem that fuzzy kernel speaker voice recognition method sensitive to outlier and noise as well as slow training, a Kernel-function based Possibilistic Fuzzy Vector Quantization (KPFVQ) was proposed. The method combines typical possibilistic clustering of fuzzy C-mean, thus suppressing the sensitivity. It also uses kernel mapping for vector quantization and match decision on voice features in the high-dimensional feature space. The characteristic differences among different samples were emphasized so that it is easier to distinguish voice and voice, voice and noise. The experiment results show that the proposed algorithm in the paper can achieve better recognition effect both to relatively clean voice and that with noise. Its training speed improves greatly as the voice length increases, which can achieve the real-time effect. 
 

Index Terms—kernel method, fuzzy C-mean, vector quantization, speaker voice recognition

Cite: Jie Yang, "Towards a Speaker Voice Recognition Method Based on KPFVQ," International Journal of Signal Processing Systems, Vol. 4, No. 5, pp. 422-426, October 2016. doi: 10.18178/ijsps.4.5.422-426