摘要
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.
作者
余本国
郇晋侠
刘晓峰
高伟涛
YU Benguo;HUAN Jinxia;LIU Xiaofeng;GAO Weitao(School of Software,North University of China,Taiyuan 030051;School of Medical Information,Hainan Medical College,Haikou 571199)
出处
《计算机与数字工程》
2021年第10期2168-2173,共6页
Computer & Digital Engineering
关键词
山西方言
语音库
语音识别
隐马尔可夫模型
Shanxi local language
speech library
speech recognition
hidden Markov model
