摘要
近年来汉语方言自动辨识研究有了初步进展,但由于缺乏带有语音标注的方言音库,性能优越的并行音素识别-语言模型(PPRLM)方法尚未得到研究和运用。本文借助高斯混合模型(GMM)符号化器把PPRLM的思想方法引入到汉语方言辨识中,并通过融合置信判别使系统能够用于开集辨识。仿真实验表明,本文方法具有很高的稳定性和可靠性,综合性能较为优越。
Lately the study of Chinese dialect identification （CDI） shows some progress. Yet the excellent method-parallel phone recognizers followed by language modeling（PPRLM）-has not be study in CDI field due to the lack of dialect corpus with annotation. In this paper, we study CDI using a method like PPRLM by virtue of GMM tokenizer, further we study the combination of a confidence measure to use the method in open-set task. Simulation results show that this CDI method is an excellent method with high stability and reliability.
出处
《计算机科学》
CSCD
北大核心
2006年第11期210-211,236,共3页
Computer Science
基金
国家社会科学基金重点项目(01AYY004)
江苏省社会科学基金项目(06J5BYY006)
江苏省"十五"社科基金项目(K3-013)
徐州师范大学人文社会科学基金项目(06XWB28)
