基于层次化结构的语言模型单元集优化

发布时间：2018-04-19 06:35

本文选题：语音识别 + 语言模型　；参考：《清华大学学报(自然科学版)》2017年03期

【摘要】：对于大词汇量语音识别系统,适当选择基本单元至关重要。虽然以词为基本单元时避免了词边界的确定等复杂过程,但很多派生类结构中(如黏性语言),词比较长,而且很多文字(如中文、日文等)不需要词边界,因而在自然语言处理应用中没有选取基本单元集的固定模式。该文以维吾尔语大词汇量语音识别系统为例,研究基于各个层次化粒度单元的语音识别系统。通过比较各种层次化单元集为基础的语音识别结果,分析错误识别模式,收集被误判的单元序列作为在2层单元序列结构中择优的训练样本库。比较各种单元集的优缺点,提出一种能平衡长单元集和短单元集优点的方法。实验结果表明:该方法不仅可以有效提高语音识别准确率,也大大缩减了词典容量。
[Abstract]:For large vocabulary speech recognition system, it is very important to select appropriate basic units.Although the word is used as the basic unit to avoid complex processes such as word boundary determination, many derived structures (such as viscous language) have longer words, and many words (such as Chinese, Japanese, etc.) do not need word boundaries.Therefore, the fixed pattern of basic unit set is not selected in natural language processing applications.Taking Uygur large vocabulary speech recognition system as an example, this paper studies a speech recognition system based on hierarchical granularity units.By comparing the results of speech recognition based on various hierarchical unit sets, the error recognition patterns are analyzed, and the misjudged unit sequences are collected as a training sample base to select the best in the two-layer unit sequence structure.By comparing the advantages and disadvantages of various cell sets, a method is proposed to balance the advantages of long and short cell sets.The experimental results show that this method can not only improve the accuracy of speech recognition, but also greatly reduce the dictionary capacity.
【作者单位】：新疆大学科学与技术学院;新疆大学信息科学与工程学院;
【基金】：国家自然科学基金资助项目(61462085,61662078,61163032) 教育部新世纪优秀人才支持计划资助项目(NCET-10-0969) 新疆维吾尔自治区高新技术发展研究计划项目(201312103)
【分类号】：TN912.34

【相似文献】