论文部分内容阅读
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。针对这一问题,提出了一种在标准维吾尔语发音字典的基础上生成多发音字典的方法。采用基于专家经验和数据驱动相结合的方法分析了维吾尔语方言口音发音变异规则,构造发音变异集合,生成初始的多发音字典,并运用了自动数据处理算法和门限阈值法,使得能够从方言口音训练语音数据中自动获得精简的多发音字典。实验结果表明:该方法对维吾尔语方言口音的识别性能有提升作用。
The standard speech-based recognition system has a lower recognition rate when recognizing speech corpus with phonetic variations. To solve this problem, a method of generating multi-pronunciation dictionary based on standard Uyghur pronunciation dictionary is proposed. Based on the combination of expert experience and data-driven method, this paper analyzes the variation rules of the pronunciation of the Uyghur dialect, constructs the variation set of the pronunciation, generates the initial polyphony dictionary, and applies the automatic data processing algorithm and the threshold threshold method to make the dialect accent Automatically obtain streamlined multi-pronunciations dictionary for training speech data. Experimental results show that this method can enhance the recognition performance of Uyghur dialect accent.