论文部分内容阅读
维吾尔语单词连接构形词缀时,经常发生元音弱化成央音的现象。但对已有形态变化的单词进行形态还原时,使用规则识别弱化央音的原音的效率一般在40%左右。提出基于噪声信道的维吾尔语央音原音识别模型。该模型以弱化词干词尾的二字符、三字符和最后音节作为上下文,建立语言模型和似然度计算公式。在开放测试中,模型的准确率达到82.45%,提高词干提取准确率15%。
When Uyghur words connect conformational affixes, the vowel weakened into the central sound often occurs. However, the morphological reduction of existing morphological words, the use of rules to identify the weakness of the original sound of the central tone efficiency is generally about 40%. A Uyghur speech recognition system based on noise channel is proposed. The model uses the two characters, the three characters and the last syllable as the context to weaken the stem ending, and establishes the language model and the likelihood calculation formula. In the open test, the accuracy of the model reached 82.45%, and the accuracy of stem extraction was improved by 15%.