科尔沁土语中蒙汉语码转换的正确标注法  

Correct Annotation Method for Mongolian-Chinese Code-Switching in the Khorchin Dialect

作　　者：白斯勤 Bai Tsetsen(School of Mongolian Studies,Inner Mongolia University;Tongliao City Bureau of Commerce,028000)

机构地区：[1]内蒙古大学蒙古学学院 [2]通辽市商务局,028000

出　　处：《中国蒙古学(蒙文)》2025年第1期28-31,224,225,共6页CHINA MONGOLIAN STUDIES

基　　金：国家社会科学基金重点项目“蒙汉混合语码口语语料库的构建及应用研究”(编号:21AZD141)。

摘　　要：蒙汉语码转换语音识别研究是蒙古语语音识别研究中难度最大的课题。研究者利用语音识别工具kaldi和点对点的语音识别方法及采用RNN、LSTM,Transformer等深度感知网络,深入研究蒙古语语音识别技术。对于使用少量数据集进行识音模型训练时Kaldi工具的效果优于点对点语音识别方法。“发音词典”帮助声学模型更准确地识别语音信号中的单词,优化语音识别能力。在转写科尔沁土语资料的过程中,为了提高“发音词典”的质量,我们不分蒙古语或汉语,将每个语音和音素——对应,准确标注了模糊元音、语音长短、前化元音、双元音等。The study of Mongolian一Chinese code-switching in speech recognition is one of the most challenging topics in Mongolic phonetic recognition research.Researchers have utilized speech recognition tools such as Kaldi,point-to-point speech recognition methods,and deep learning networks like RNN,LSTM,and Transformer to delve into Mongolian speech recognition technologies.When training speech models with small datasets,Kaldi outperforms point-to-point recognition methods.The"pronunciation dictionary"helps the acoustic model more accurately identify words in speech signals,thus optimizing recognition capability.In transcribing Khorchin dialect materials,to improve the quality of the"pronunciation dictionary,"each speech signal and its phoneme were meticulously aligned,without distinguishing between Mongolian and Chinese,and accurately annotated for features such as ambiguous vowels,vowel length,front vowels,and diphthongs.

关 键 词：蒙汉语 语码转换 标注法 

分 类 号：H31[语言文字—英语]