摘要： 用独立分量分析的方法计算每一种待识别语言的特征向量空间的基函数组及其系数向量各分量的概率分布，并用这两组信息来惟一刻画一种语言。测试音频通过上述两组信息计算针对每一种语言的后验概率，具有最大后验概率的语言就是最终的识别结果。实验结果表明，该方法具有快速、高效的特点。

关键词: 独立分量分析, 基函数组, 系数向量, 后验概率

Abstract: The performance of language identification depends heavily on the description of differences between languages. This paper exclusively describes one language through two sets of data—the basis functions of the feature space and the probability distribution of every dimension of coefficient vector, which are calculated using independent component analysis method on its training data. A posterior probability is computed through the two sets of information when a match between the test speech and one specific language is evaluated. The language corresponding to the maximum posterior probability serves as the identification result. The algorithm is proven to be fast and effective by the result of experiments.

Key words: Independent component analysis, Basis functions, Coefficient vector, Posterior probability