基于改进的i-vector 的方言语种识别

Dialect Language Recognition Based on Improved i-vector

经典的i-vector的提取方法利用方言特征在通用背景模型(Universal Background Model,UBM)的统计差异来构建全局差异空间,对方言语种的区分能力较弱。为此,提出了一种基于改进的i-vector的提取算法,利用方言特征在方言相关的高斯混合模型(Gaussian Mixture Model,GMM)上的统计差异来构建全局差异空间,提升i-vector对方言语种的区分能力。首先基于方言相关GMM分别构建全局差异空间;其次拼接各空间中提取到的i-vector并进行主成分分析(Principal Component Analysis,PCA)降维,得到改进的i-vector;最后采用高斯概率线性判别分析(Gaussian Probabilistic Linear Discriminant Analysis,GPLDA)模型进行建模和打分。实验表明,所提算法较经典i-vector算法能更有效地提升对方言语种的识别性能。

In classical i-vector extraction algorithm,a global variability space is constructed using the statistical differences of dialect features based on UBM(Universal Background Model),which is weak in distinguishing among dialect varieties.To address this problem,this paper proposes an extraction algorithm based on an improved i-vector.The method uses the statistical differences of dialect features on dialectrelated GMM(Gaussian Mixture Model)to construct a global variability space,which improves the ability of the i-vector to discriminate among dialect varieties.First,global variability spaces are constructed separately based on dialect-related GMMs.Then,the improved i-vectors are obtained by combining the i-vectors extracted from each global variability space and performing PCA(Principal Component Analysis)and dimensionality reduction.Finally,the GPLDA(Gaussian Probabilistic Linear Discriminant Analysis)is used for modeling and scoring.The experimental results indicate that the proposed algorithm outperforms the classical i-vector algorithm in dialect language recognition.