摘要
为了解决老年人网上购票困难的问题,通过智能语音识别技术的运用,实现铁路12306手机客户端语音购票功能,降低老年人网上购票的操作门槛。目前语音识别算法的准确率已经达到较高水平,但是通用的语音识别方法存在铁路专有名词识别困难的问题。针对此,提出融合铁路专有知识的语音识别方法,使用多重语言模型共同修正声学模型的解码结果;增加前缀束搜索的解码方法,提高解码过程中的召回率;设计的热词赋权模块,配置包含铁路专有名词的热词库,提高专有名词的检出率。在自建数据集上进行对比实验,结果表明,提出的基于多重语言模型融合的铁路购票语音识别方法能准确识别铁路专有名词,达到90%以上的识别准确率。
To solve the difficulty of online ticket purchase for the elderly, we lower the operation threshold by implementing a voice-operated ticket purchase function on the 12306 mobile clients with intelligent speech recognition technology. Although the accuracy of speech recognition algorithms has reached a high level, general speech recognition methods have difficulties in recognizing railway terms. A speech recognition method that integrates railway-specific knowledge was proposed to solve this problem. Specifically, multiple language models were used to modify the decoding results of the acoustic model. The decoding method of prefix beam search was added to increase the recall rate in the decoding process. The hot word weighting module designed was equipped with a hot word library containing railway terms to improve the detection rate of such terms. Comparative experiments were carried out on the self-built dataset.The results show that the proposed speech recognition method for railway ticket purchase based on the fusion of multiple language models can accurately recognize railway terms, with the accuracy reaching more than 90%.
作者
王心雨
单杏花
景辉
WANG Xinyu;SHAN Xinghua;JING Hui(Postgraduate Department,China Academy of Railway Sciences,Beijing 100081,China;Institute of Computing Technology,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)
出处
《铁道运输与经济》
北大核心
2022年第3期23-30,共8页
Railway Transport and Economy
基金
国家重点研发计划项目(2020YFF0304101)。
关键词
语音购票
语音识别
多重语言模型
铁路专有名词
热词
Voice-operated Ticket Purchase
Speech Recognition
Multiple Language Models
Railway Terms
Hot Words
