摘要
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。
In this paper, HTK （Hidden Markov model-based Toolkit） based Uyghur continuous phoneme recognition baseline system is presented, and its several language-depended key technologies are addressed. According to the characteristics of Uyghur language, it designs the text corpus for language modeling and speech corpus construction, and records a large-scale speech data for training the phoneme based Uyghur acoustic model. The different recognition rates with different N-gram language models are also given. The statistics of the recognition rates of 32 Uyghur phonemes, the list of the confused phonemes and their possible reasons are analyzed. And then it gives some research directions for further improvements to the baseline system.
出处
《计算机工程与应用》
CSCD
2013年第22期150-154,172,共6页
Computer Engineering and Applications
基金
国家自然科学基金(No.61163032
No.61065005)
