Text-Independent Speaker Age Recognition Based On Feature Sub-Space Quantization

Speaker recognition is the identification of the age characteristics of the speaker by speech signal analysis.With the continuous development of human computer interaction technology,its application is more and more extensive.Speaker recognition technology can make the system age correctly understand the age characteristics of the speaker in the process of human-computer interaction,so as to provide a reasonable way of adaptive interaction,such as proper volume,speed,intonation,tone,etc..The technology can be widely used in the field of automatic voice information query,health care,entertainment and other fields.This paper proposes a feature subspace quantization(FSSQ,Feature Subspace Quantization)scheme for text independent speaker recognition age,the main idea is the distribution of divergence through acoustic feature space of the same age speaker clustering technology based on Subspace Partition and subspace quantization to reduce the pattern,improve the overall recognition accuracy.Speech signal is the same age of the first speaker Mel cepstrum extraction(MFCC),and then use K-Means algorithm to cluster the feature vectors,feature subspace partition,further uses the LBG algorithm to quantify each subspace,the formation of the codebook,speech each age finally expressed as a set of quantization code this.Age recognition based on the minimum average code of this distance for decision classification.The experimental results show that the subspace quantization speaker age recognition method proposed relative vector quantization(VQ)and Gauss mixture model(GMM)method has better recognition performance and other typical,set and set the overall recognition rate reached 89.8% and 58.6%.