基于说话人聚类和高斯混合模型的语言辨识研究

Language Identification Using Speaker Clustering and Gaussian Mixed Model

本文给出了一种语言辨识的新方法。通常来讲,语言辨识系统是说话人无关的,但说话人的个体特征对语言辨识系统有很大的影响,文本采用了一种粗分类精识别的思想,利用说话人聚类技术有效解决了粗分类的问题,对每类相近说话人集合建立模型,然后进行识别。实验表明,该方法对于说话人无关的语言辨识问题是有效的。

In this paper, a novel approach to language identification is proposed. Generally speaking, ideal automatic language identification system is speaker-independent, but the personal characteristics of speakers have an important influence on the performance of language identification systems. Here, we utilize an idea of rough classification and fine recognition, namely, the rough classification is realized by using speaker clustering, then the models are constructed based on each subset of speakers so as to perform fine recognition. Preliminary results on language identification are provided to demonstrate the effectiveness of such system.