论文部分内容阅读
以建立维吾尔语连续音素识别基础平台为目标,在HTK(基于隐马尔可夫模型的工具箱)的基础上,首次研究了其语言相关环节的几项关键技术;结合维吾尔语的语言特征,完成了用于语言模型建立和语音语料库建设的维吾尔语基础文本设计;根据具体技术指标,录制了较大规模语音语料库;确定音素作为基元,训练了维吾尔语声学模型;在基于字母的N-gram语言模型下,得出了从语音句子向字母序列句子的识别结果;统计了维吾尔语32个音素的识别率,给出了容易混淆的音素及其根源分析,为进一步提高识别率奠定了基础。
Based on HTK (Hidden Markov Model Toolbox), this paper firstly studies several key technologies in the language related aspects of Uyghur language. Based on the linguistic features of Uyghur language, The basic Uyghur text design for language model building and phonetic corpus construction was recorded. According to specific technical indicators, a larger scale corpus of phonetics was recorded. The phoneme was defined as a primitive to train the Uyghur language acoustics model. Based on the letter N-gram Linguistic model, the result of recognizing sentences from phonetic sentences to alphabetic sequences is obtained. The recognition rate of 32 phonemes in Uyghur language is calculated, the easily confused phonemes are analyzed and their root causes are analyzed, which lays the foundation for further improving the recognition rate.