论文部分内容阅读
为提高维吾尔语语音识别的识别率,在分析维吾尔语特点的基础上,设计一种基于子字单元的维吾尔语语音识别总体结构,指出维吾尔语单词的发音模型,给出构建子字发音字典的方法,及其以子字单元为基础构建语言模型与声学模型的方法。在一个语音库上进行实验,采用一种非监督的词切分方法对维吾尔语单词进行词切分,生成子字。实验结果表明,基于子字单元的维吾尔语语音识别可以获得更好的识别结果。
In order to improve the recognition rate of Uyghur speech recognition, based on the analysis of the Uyghur features, this paper designs a Uyghur speech recognition general structure based on sub-word units, points out the pronunciation model of Uyghur words, Method, and the method of constructing language model and acoustic model based on sub-word unit. Experiments were conducted on a phonetic database, using an unsupervised word segmentation method to segment the Uyghur words and generate subwords. Experimental results show that Uyghur speech recognition based on sub-word unit can obtain better recognition results.