面向电力巡检语音指令识别后的文本纠错算法  

Text Error Correction Algorithm After Speech Command Recognition for Power Inspection

作　　者：陈楠 曹雪虹[1] 焦良葆[1] 孟琳 徐逸 张四维 CHEN Nan;CAO Xuehong;JIAO Liangbao;MENG Lin;XU Yi;ZHANG Siwei(Institute of Artificial Intelligence Industry Technology,Nanjing Institute of Technology,Nanjing 211167;State Grid Nanjing Power Supply Company,Nanjing 210009)

机构地区：[1]南京工程学院人工智能产业技术研究院,南京211167 [2]国网南京供电公司,南京210009

出　　处：《计算机与数字工程》2022年第1期116-123,134,共9页Computer & Digital Engineering

基　　金：国家自然科学基金青年基金项目(编号:61903183)资助。

摘　　要：为了解决面向公共领域的语音识别(Automatic Speech Recognition,ASR)在识别电力巡检专有指令时识别率低的问题,提出了基于模型匹配的电力巡检语音识别后文本纠错算法。首先对庞大的电力巡检指令词库进行预处理并构造电力语法模型,建立电力词汇间的联系,然后对该模型下的电力词汇构建汉字和拼音AC自动机(Aho-Corasick Automaton,ACA)分别进行识别串精确匹配下的查错和纠错,最后针对精确匹配下纠错失败的问题设计模糊匹配算法,进一步实现语音识别后文本的模糊纠错,从而优化了电力巡检语音识别后文本纠错的性能。实验结果表明,在满足实际的应用需求的情况下,该算法解决了电力巡检语音指令词的误识别问题,显著提高了纠错正确率。Aiming at the problem of low recognition rate of automatic speech recognition(ASR)for the public domain in recog⁃nizing exclusive instructions for power inspection,a text error correction algorithm based on model matching is proposed for power inspection after speech recognition.First,the huge power inspection instruction lexicon is preprocessed and the power grammar mod⁃el is constructed to establish connections between power vocabularies.Second,Chinese characters and pinyin Aho-Corasick automa⁃ton(ACA)are built under the model for power vocabulary to perform recognition string exact matching checks.Finally,the fuzzy matching algorithm is designed for the problem of error correction failure under precise matching,which further realizes the fuzzy er⁃ror correction of the text after speech recognition,thereby optimizing the text error correction performance after speech recognition in the power inspection.The results show that the proposed algorithm can solve the problem of misrecognition of speech command words in power inspection and significantly improves the error correction rate while meeting actual application requirements.

关 键 词：语音识别 电力语法模型 AC自动机 精确匹配 模糊匹配 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]