维吾尔语语音识别中发音变异现象

Uyghur pronunciation variations in automatic speech recognition systems

维语口语发音中很多音素相对标准语产生了发音变异,基于标准语音的识别系统在识别带有发音变异的口语语料时识别率较低。该文针对维吾尔语同化、弱化、脱落、元音和谐等语流音变难点进行分析,对语音、韵律特性进行知识融合与技术创新,运用基于数据驱动和基于专家经验相结合的方法对维吾尔语方言口语中存在的发音变异现象进行研究,统计元音、辅音多发音变化映射对,建立音素混淆矩阵,为维吾尔语方言口语语音识别研究奠定基础。

The recognition rate in Uyghur standard speech recognition systems is relatively low due to pronunciation variations in the corpus.The assimilation,weakening,loss,and vowel harmony in Uyghur speech were analyzed to provide a better understanding of the voice and rhythm characteristics.The Uyghur accent pronunciation variation rules are summarized by combining knowledge-based and data-driven methods with mapping of vowel and consonant pairs and developing of a phoneme confusion matrix.