语音识别系统在山西方言中的实现与应用

Implementation and Application of Speech Recognition System in Shanxi Dialect

目前山西的语音识别系统多数为普通话识别,对于该地区方言识别的准确率并不理想。针对这一问题,采集山西地方方言语音和语料建立语音库,根据山西各地方言发音的特点,构建山西地方方言的语音识别系统,以山西声韵母为基元,提取Mel倒谱系数(MFCC)的特征参数,选择隐马尔可夫模型(Hidden Markov Model,HMM),实现山西当地方言的语音识别系统。实验结果显示,针对差别小的小区域方言识别,HMM的识别率有很好的稳定性。

At present,most of the speech recognition systems in Shanxi are mandarin recognition,which is not ideal for the accuracy of dialect recognition in the region.In response to this problem,Shanxi local dialect speech and corpus are collected to establish a speech database.Based on the pronunciation characteristics of Shanxi local dialects,a Shanxi local dialect speech recognition system is constructed.Based on the Shanxi vowels,the characteristic parameters of Mel cepstrum coefficient(MFCC)are extracted,hidden Markov model(HMM)is selected to realize the speech recognition system of Shanxi local dialect.Experimental results show that the recognition rate of HMM has good stability for small regional dialect recognition with small differences.