基于音素识别的语种辨识方法中的因子分析

Factor Analysis for Language Identification Based on Phoneme Recognition

在基于音素识别的语种辨识系统中,特定的一段语音,音素识别的结果会受到说话人和信道等干扰因素的影响.对此,文中基于音素搭配关系对每段语音构建相应的特征向量表示.在向量空间中,利用因子分析建立噪声子空间的数学描述模型,并在语言模型的训练和识别过程加以消除.在NISTLRE2007的测试任务中,相对于基于音素识别的语种辨识基线系统,该方法可有效提高系统性能.在30s时长测试中,基于音素识别的语言模型和基于音素识别的支持向量机模型的等错误率分别相对降低14.4%和12.9%.

In the phoneme recognition based language identification system,the key issue is whether the tokens or the token sequence can reflect the language related information or not.However,it is observed that for certain utterance,the noise in the output token sequence from the phone recognizer is introduced due to the channel,speaker and background clutters.To address this problem,each utterance is represented in n-gram vector.And in this vector space,the factor analysis is applied to model the noise subspace,which will be reduced in final modeling process.The experiment results on NIST LRE 2007 show that the proposed method can outperform the existing phone recognition based language identification system.In 30s evaluation task,the equal error rate（EER） of recognition reduces relatively about 14.4% against the baseline phone recognition followed by language modeling（PRLM） system,while about 12.9% against the baseline phone recognition followed by support vector machine（PRSVM） system.