中文摘要：

探索一种从语流中自动提取伪音节的新方法，该方法可以用于自动语种识别（ALI）．整个过程分为特征提取、模型建立和识别测试3个阶段．为了从语流中自动提取伪音节，将紧邻的一个辅音段和一个元音段结合在一起构成一个伪音节，并称之为CV音节．提出了一种自动提取CV音节的算法，利用该算法可以提取出每个CV音节的特征矢量．采用高斯混合模型（GMM）和语言模型（LM）构建语种识别系统．对汉语普通话及6种少数民族语言的实验证明了提出的方法能够有效地识别语种，而且训练速度快、抗噪声性能强．

英文摘要：

A new method of automatically extracting pseudo-syllable from a flow of speech is explored. The method can be applied to automatic language identification （ALl）. The whole procedure includes three phases： feature extraction, mod- eling and identification test. In order to automatically extract pseudo-syllable from a flow of speech, a consonant segment is integrated with a vowel segment to form a pseudo-syllable which is called CV-syllable. Moreover, an algorithm of automatic CV-syllable extraction is proposed. By using the algorithm, a feature vector can be extracted from each CV-syllable. The Gaussian mixture model （GMM） and language model （LM） are employed to build a language identification system. Experi- ments on Mandarin and six minority languages prove that the proposed method can effectively identify languages, and that it is fast in the training process and robust against the noise.