IMUT-MC:一个针对蒙古语语音识别的语音语料库  

IMUT-MC:a speech corpus for Mongolian speech recognition

作　　者：刘志强 马志强 张晓旭 宝财吉拉呼 谢秀兰 朱方圆 LIU Zhiqiang;MA Zhiqiang;ZHANG Xiaoxu;BAO Caijilahu;XIE Xiulan;ZHU Fangyuan(College of data science and Application,Inner Mongolia University of Technology,Huhhot,P.R.China;Inner Mongolia Autonomous Region Software Service Engineering Technology Research Center Based on Big Data,Huhhot,P.R.China)

机构地区：[1]内蒙古工业大学数据科学与应用学院,呼和浩特010000 [2]内蒙古自治区基于大数据的软件服务工程技术研究中心,呼和浩特010000

出　　处：《中国科学数据（中英文网络版）》2022年第2期71-83,共13页China Scientific Data

基　　金：国家自然科学基金(61762070,62166029);内蒙古自然科学基金(2019MS06004)

摘　　要：蒙古语作为少数民族语言,其使用人群分布辽阔,收集标注语音数据困难,导致没有公开的大规模蒙古语语音语料库为广大研究人员提供实验支撑,阻碍了蒙古语语音识别的进一步发展。本课题组构建了一个针对蒙古语语音识别任务的语音语料库IMUT-MC,包含417位说话人录制的约212小时的阅读语音,致力于推进蒙古语语音识别研究。课题组分别在传统语音识别模型和端到端语音识别模型上使用IMUT-MC进行基线语音识别实验,基于GMM-HMM、DNN-HMM和Transformer的语音识别模型在IMUT-MC上词错率分别为69.90%、67.45%和26.10%,证明了IMUT-MC是进行蒙古语语音识别可靠的语料库。There is a lack of large-scale speech corpora of Mongolian(a minority language)accessible to researchers for experimental reference,because its users are scattered and it is difficult to collect and label the speech sounds,which hinders the further development of Mongolian speech recognition.Our research group has constructed a speech corpus IMUT-MC for Mongolian speech recognition tasks,which contains about 212 hours of reading speech recorded by 417 spea kers,and we are committed to advancing Mongolian speech recognition research.The research group used IMUT-MC to conduct baseline speech recognition experiments on traditional speech recognition models and end-to-end speech recognition models respectively.The speech recognition models based on GMM-HMM,DNN-HMM and Transformer have word error rates on IMUT-MC,respectively.69.90%,67.45%and 26.10%,which proves that IMUT-MC is a reliable corpus for Mongolian speech recognition.

关 键 词：蒙古语 语音识别 语音语料库 阅读语音 

分 类 号：H212[语言文字—少数民族语言] TN912.34[电子电信—通信与信息系统]