基于三音子模型的柯尔克孜最优语料选取算法

Kyrgyz's Optimal Corpus Selection Algorithm Based on Triphone Model

选择具有丰富语音现象的语料库是提高语音识别性能的关键。为了构建柯尔克孜语语音识别文本语料库,首先利用预处理技术去除文本中的噪声信息并用文本转换算法将柯尔克孜文转换为拉丁文形式。其次,根据柯尔克孜语的音节结构和规则,提出了启发函数和两种最优自动选择句子的算法。最后,为了验证算法的有效性,将两组包含不同数量的句子集作为实验语料,采用两种算法生成最优句子集,并对两种算法生成的语料库进行了统计,实验结果表明,利用算法2挑选出来的文本包含的三音子覆盖率达到了78.70%,能够满足语音识别系统的需要,验证了提出的算法的有效性。

Choosing a corpus with rich phonetic phenomena is the key to improve the performance of speech recognition.In order to construct the text corpus of Kyrgyz speech recognition system,firstly,the noise information in the text is removed by pre-processing technology,and the Kyrgyz language is converted into Latin form by text conversion algorithm.Secondly,according to the syllable structure and rules of Kyrgyz language,the heuristic function and two optimal algorithms for automatically selecting sentences are proposed.Finally,in order to verify the effectiveness of the algorithm,two groups of sentence sets with different numbers are used as experimental corpora,two algorithms are used to generate the optimal sentence sets,and the corpora generated by the two algorithms are counted.The experimental results show that the coverage rate of tri-phones in the text selected by algorithm 2 reaches 78.70%,which can meet the needs of speech recognition system,and the effectiveness of the algorithm proposed in this paper is verified.