面向语音合成的维吾尔语音素自动切分算法研究  

ON SPEECH SYNTHESIS-ORIENTED AUTOMATIC SEGMENTATION ALGORITHM OF UYGHUR PHONEME

作　　者：阿依木尼萨.胡甫尔 艾斯卡尔.艾木都拉[1] 

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《计算机应用与软件》2011年第9期18-21,共4页Computer Applications and Software

基　　金：国家自然科学基金(61062008);国家语委项目(MZ115-75)

摘　　要：结合维吾尔语语音特征,以建立维吾尔音素语料库为目标,为了减少人工工作量,通过HTK工具实现了音素的自动切分算法:首先完成了文本设计、录音和手动标注等准备工作,设计了上下文属性集,通过训练获得了每个音素的HMM模型,随后对任意输入的语音句子进行了其音素构成部分的自动切分,最后分析了其切分准确度、存在的问题及对策等。实践表明,在语料库的建设中,该研究策略确实节省了大量的时间和人力成本,提高了语音语料库标注信息的一致性和准确性。In combination with Uyghur phonetic features,aiming at building Uyghur phoneme corpus and for reducing manual workload,we realised the phoneme automatic segmentation algorithm with the HTK tool as presented in this paper.First,we completed the preparatory works such as text design,sound recording and manual labelling,and designed the context attribute collection,the HMM models of each phoneme was obtained through training.Then we automatically segmented the phonemic components of arbitrarily inputted vocal sentences.At last we analysed its segmentation accuracy,the problems existed and their countermeasures,etc.Practice indicated that during the construction of the corpus,the strategy studied in this paper did have saved massive time and manpower cost,and improved the consistency and the accuracy of labelled information of speech corpus.

关 键 词：音素自动切分 隐马尔科夫模型 维吾尔语 语音合成 音素语音语料库 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]