摘要
随着智能电网的高速发展,电网业务中对语音识别的需求也在不断增加。然而面向公共领域的语音识别技术很难识别出电网特有的专业信息词汇,使得电力行业语音的识别准确率不高。基于此,设计一种面向电力行业的热词语音识别技术,技术首先构建CTC声学模型将语音信息转化为基本音素信息,再利用电力行业热词库构建针对电力数据的Transformer语言模型,最后通过语言模型和发音字典将基本音素信息解码为中文信息,并通过基于南网信息系统语料库的实验验证该方法的有效性。
With the rapid development of the smart grid,the demand for voice recognition in the grid business is also increasing.However,the voice recognition technology for the public domain is difficult to recognize the professional information vocabulary unique to the power grid,which makes the voice recognition accuracy of the power industry not high.Based on this,this paper designs a hot word speech recognition technology for the power industry.The technology first builds a CTC acoustic model to convert voice information into basic phoneme infor⁃mation,then uses the power industry hot lexicon to build a Transformer language model for power data,and finally decodes the basic pho⁃neme information into Chinese information through the language model and pronunciation dictionary.The effectiveness of this method is verified by experiments based on the corpus of the State Grid Information System.
作者
张云翔
李智诚
ZHANG Yun-xiang;LI Zhi-cheng(Shenzhen Power Supply Bureau Co.,Shenzhen 518001)
出处
《现代计算机》
2020年第22期14-17,共4页
Modern Computer