摘要
目前语音识别系统以识别普通话为主,用户只有以标准普通话发音才能得到较好的识别结果。对于方言较重的用户,识别准确率下降较为明显。针对现有语音识别系统在电信客服领域中对方言识别能力差的问题,本文提出了一种基于方言区域的分省优化策略。首先研究了国内方言的分布情况,并将全国31省分为9个方言区域,然后在方言区域的基础上对区域内各省进行针对性优化。实验结果表明,优化策略在方言较严重的川、贵、渝区域的平均识别准确率提升近1. 37%。
At present,the speech recognition system is mainly used in identify Mandarin.Only when users pronounce with standard Mandarin can they get better recognition results.However,for the heavier dialect users,the accuracy rate of recognition decreased significantly.In view of the poor recognition ability of the existing speech recognition system in the field of actual telecommunication service,this paper proposes an optimization strategy for the speech recognition system with sub regional training in regional dialect.This paper studies the distribution of domestic dialects,and then divides the 31 provinces into 9 dialects firstly.And then,the individual province optimization in its regional dialect is conducted.The experiment result shows that the average recognition accuracy of Sichuan-Guizhou-Chongqing region improvement 1.37%on our optimization strategy.
作者
杨正哲
任玉玲
杜省
柳瑞波
YANG Zhengzhe;REN Yuling;DU Sheng;LIU Ruibo(China Mobile Online Service Company Limited,Zhengzhou,450001,China)
出处
《网络新媒体技术》
2019年第1期37-42,共6页
Network New Media Technology
关键词
智能客服
语音识别
方言识别
Smart customer service
Speech recognition
Dialect recognition
