语言辨识系统中最佳线性融合技术的研究

Optimal Linear Combinations of Classifiers for Language Identification

本文利用不同参数提取方法对语言辨识系统中的线性融合技术进行了研究。融合系数的获取通过三个准则进行实现,CFM准则、MSE准则和CE准则。实验系统采用了区分性高斯混合模型,利用OGI-TS多语种电话语音语料库,对决策级融合性能进行了评估。实验表明,利用决策级融合技术,选择最佳融合系数,可以很好地改善语言辨识率。

This paper presents the fusions for optimally combining different language identification using different features. The optimal combining coefficients are obtained using three criterions. The criterions considered are; classification figure of merit （ CFM ）, mean square error（MSE） and cross entropy（CE）. The reference system uses the discriminative training algorithm to get each model parameters. The experiments are conducted using OGI Multi-language speech corpus. The experimental results show the optimal combination of different classifiers using different parameters is very effective in improving the language identification accuracy rates.