论文部分内容阅读
推行新疆地区的双语教学是国家推进民族交流的重要举措,对维语进行自动发音质量评估能够大大提高双语教学的效率.然而维吾尔语作为黏着语,其特殊的构词方法造成大量无法被收入词表的集外词存在,使得基于传统语音识别系统的发音评估方法难以应用.为了实现高效的维语发音质量评估,本文在研究分析了维语的发音规则和发音习惯后,决定采用子词作为基本的识别单元;在原有发音质量自动评估系统上,改用基于双层词法分析的维语子词切分器;同时引入音素解码器计算后验概率的分母,让系统可以在子词级别直接计算置信度.经过实验数据的验证,基于子词的维语评估系统的性能要优于基于音素的系统.
Carrying out bilingual education in Xinjiang is an important measure for the government to promote ethnic communication, and the automatic pronunciation quality evaluation of Uyghur language can greatly improve the efficiency of bilingual teaching.However, Uyghur language, as a sticky language, has a large number of unacceptable income words The existence of extra words in the table makes it difficult to apply the pronunciation evaluation method based on the traditional speech recognition system.In order to evaluate the quality of the Uygur language pronunciation effectively, this paper, after studying and analyzing the pronunciation rules and pronunciation habits of Uyghur language, The basic recognition unit; in the original automatic pronunciation quality evaluation system, switched to double-dimensional lexical analysis of the word semantic word divider; also introduced the phoneme decoder to calculate the posterior probability denominator, the system can be directly in the sub-word level Calculate the confidence level.After verifying the experimental data, the performance of the sub-word-based Uygur language evaluation system is better than the phone-based system.