计算机技术与发展2018，Vol.28Issue(2)：45-49,5.DOI:10.3969/j.issn.1673-629X.2018.02.011
基于AutoEncoder DBN-VQ的说话人识别系统
Speaker Recognition System Based on AutoEncoder Deep Belief Network and Vector Quantization
摘要
Abstract
The speaker recognition system using vector quantization works by describing the different characteristics of the speaker's speech features.When the number of speakers are large and training speech length is short,the recognition rate of the system is not high.For the model is usually trained under the condition of pure speech,the performance of the system will be poor when it is used in the actual environ-ment.In order to improve the recognition performance of the system,we propose a method of speech recognition based on the combination of AutoEncoder deep belief network and vector quantization.It adopts the deep belief network to model and learn for speech data,so speaker's personality characteristics in speech can be better captured when the speech length is short.In the meantime,it structures AutoEncoder deep belief network,which is effective on noise filtering for noisy speech data.The experiment show that the proposed method can improve the recognition rate greatly when there is only a small amount of speaker training data and speech is noisy.关键词
说话人识别/深度置信网络/自动编码器/矢量量化Key words
speaker recognition/deep belief network/AutoEncoder/vector quantization分类
信息技术与安全科学引用本文复制引用
刘俊坤,李燕萍,凌云志..基于AutoEncoder DBN-VQ的说话人识别系统[J].计算机技术与发展,2018,28(2):45-49,5.基金项目
国家自然科学基金(61401227) （61401227）
江苏省博士后基金(1402067B) （1402067B）
