%0 Journal Article
%T THUYG-20:免费的维吾尔语语音数据库<br>THUYG-20: A free Uyghur speech database
%A 艾斯卡尔·肉孜
%A 殷实
%A 张之勇
%A 王东
%A 艾斯卡尔·艾木都拉
%A 郑方
%J 清华大学学报（自然科学版）
%D 2017
%R 10.16511/j.cnki.qhdxxb.2017.22.012
%X 语音数据资源是语音识别研究的基础。当前国内只有为数不多的开放的语音数据库供研究者免费使用，特别是在维吾尔语等少数民族语音识别方面，数据资源更为贫乏。该文发布一个完全免费的维吾尔语连续语音数据库，该数据库包括约20 h的训练数据和1 h的测试数据，同时介绍了构建维吾尔语语音识别系统所需要的音素集、词表、文本数据等相关资源，以及用于构建基线系统的脚本。给出了该基线系统在纯净测试数据和噪声测试数据上的识别性能。该数据库为维吾尔语语音识别研究提供了可以借鉴的标准数据库。<br>Abstract：Speech data plays a fundamental role in research on speech recognition. However, there are few open speech databases available for researchers in China, especially for minor languages such as Uyghur. This paper develops a Uyghur continuous speech database which is totally open and free. The database consists of 20 h of training speech and 1 h of test speech, as well as all the resources needed to construct a full Uyghur speech recognition system, including a phone set, lexicon, and text data. A recipe used to construct the baseline system is also described with results for two test sets involving clean speech and noisy speech. This paper provides a standard database for Uyghur speech recognition.
%K 语音识别
%K 维吾尔语
%K 语料库
%K 深度神经网络(DNN)
%K &lt
%K br&gt
%K speech recognition
%K Uyghur language
%K corpus
%K deep neural network (DNN)
%U http://jst.tsinghuajournals.com/CN/Y2017/V57/I2/182