摘要
针对目前语音交互技术语音识别效果较差的问题,文中基于Viterbi搜索解码技术,结合前端处理(语音信号)、声学和语言模型,在连续语音识别结构中引入了音素、音节识别结构,设计了面向汉字关键词的智能语音交互算法。该算法依次通过连续语音识别器、关键词搜索器、置信度确认、关键词确认等过程,实现了对语音中关键词的有效检测。通过在人机交互平台的测试结果表明,该算法的查准率为90%、召回率为95%、误识率为13%,具有良好的性能。
According to the poor speech recognition effect of the current speech interaction technology,this paper proposes an intelligent voice interaction algorithm for Chinese keywords based on the Viterbi search decoding technology,combined with the front⁃end processing(speech signals),acoustics and language models.The phonemes and syllable recognition structures are introduced into the continuous speech recognition structure in the algorithm.The operation process of the algorithm includes the continuous speech recognizer,keyword searcher,confidence confirmation,keyword confirmation,achieving effective detection of keywords in speech.After long⁃term testing of the human⁃computer interaction platform,it is found that the algorithm has a high accuracy rate 90%,recall rate 95%and a low misunderstanding rate 13%,which can provide a reference for the design of the related algorithms.
作者
黄小奇
范晟
陈光文
许卓伟
彭锴
方志丹
王烁
HUANG Xiaoqi;FAN Sheng;CHEN Guangwen;XU Zhuowei;PENG Kai;FANG Zhidan;WANG Shuo(Shantou Power Supply Bureau,Guangdong Power Grid Co.,Ltd.,Shantou 515000,China)
出处
《电子设计工程》
2021年第10期37-41,共5页
Electronic Design Engineering
基金
广东电网信息管理智能化(0305002018030303XX00070)。
