论文部分内容阅读
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。
For large vocabulary speech recognition systems, the proper choice of basic unit is crucial. Although the word-based unit avoids the complicated process of word boundary determination, many derived class structures (such as sticky language) have relatively long words and many words (such as Chinese, Japanese, etc.) do not need word boundaries, thus There is no fixed pattern for selecting basic unit sets in natural language processing applications. This paper takes Uyghur large vocabulary speech recognition system as an example to study the speech recognition system based on each hierarchical granularity unit. By comparing the results of speech recognition based on various hierarchical unit sets, the pattern of misidentification was analyzed, and the misjudged unit sequence was collected as a training sample base which is optimal in the sequence structure of 2-level unit. Comparing the advantages and disadvantages of various unit sets, a method that can balance the advantages of long unit sets and short unit sets is proposed. Experimental results show that this method can not only improve the accuracy of speech recognition, but also greatly reduce the dictionary capacity.