基于稀疏组lasso约束的本征音子说话人自适应.pdfVIP

第36 卷第9 期 通 信 学 报 Vol.36 No.9 2015 年9 月 Journal on Communications September 2015 doi:10.11959/j.issn.1000-436x.2015241 基于稀疏组 LASSO 约束的本征音子说话人自适应 屈丹，张文林 （信息工程大学 信息系统工程学院，河南 郑州 450000 ） 摘 要：本征音子说话人自适应方法在自适应数据量不足时会出现严重的过拟合现象，提出了一种基于稀疏组 LASS 约束的本征音子说话人自适应算法。首先给出隐马尔可夫—高斯混合模型下本征音子说话人自适应的基 本原理；然后将稀疏组 LASS 正则化引入到本征音子说话人自适应，通过调整权重因子控制模型的复杂度，并 通过一种加速近点梯度的数学优化算法来实现；最后将稀疏组 LASS 约束的自适应算法与当前多种正则化约束 的自适应方法进行比较。汉语连续语音识别的说话人自适应实验表明，引入稀疏组LASS 约束后，本征音子说 话人自适应方法的性能得到了明显提高，且稀疏组LASS 约束方法优于l1、l2 和弹性网正则化方法。 关键词：说话人自适应；本征音子；组稀疏约束；稀疏组LASS 约束；近点梯度法 中图分类号：TN912.34 文献标识码：A Sparse group LASSO constraint eigenphone speaker adaptation method for speech recognition QU Dan, ZHANG Wen-lin (Institute of Information System Engineering, PLA Information Engineering University, Zhengzhou 450000, China) Abstract: Original eigenphone speaker adaptation method performed well when the amount of adaptation data was suffi- cient. However, it suffered from server overfitting when insufficient amount of adaptation data was provided. A sparse group LASSO(SGL) constraint eigenphone speaker adaptation method was proposed. Firstly, the principle of eigenphone speaker adaptation was introduced in case of hidden Markov model-Gaussian mixture model (HMM-GMM) based speech recognition system. Then, a sparse group LASSO was applied to estimation