摘要
针对当前会议记录报告智能生成方法普遍存在的错记和漏记问题,提出一种基于语音识别技术的会议记录报告智能生成方法。针对发言人的语音信号进行去噪和预加重处理,从预处理后的语音信号中提取FBank特征和MFCC特征。基于双层神经网络构建语音识别模型,以语音特征为输入,在识别发言人身份的同时识别发言内容,该模型可将发言内容实时转写为文字,并自动整理成会议记录报告,实现会议记录报告的智能生成。实验结果表明,所研究方法的漏记率和错记率均较低,说明该方法生成的会议记录更完整、错误率更低,更能保证会议记录质量。
Aiming at the common problems of misrecording and missing recording in the intelligent generation method of meeting minutes report,this paper studies an intelligent generation method of meeting minutes report based on speech recognition technology.The speaker’s speech signal is denoised and pre⁃stressed,and FBank and MFCC features are extracted from the pre⁃processed speech signal.The speech recognition model is built based on the two⁃layer neural network,and the speech content is recognized with the speech features as input.The model can transcribe speech content into text in real time and automatically organize and generate meeting record reports,achieving intelligent generation of the meeting record report.The experimental results show that the omission rate and error rate of the proposed method are both low,which indicates that the meeting minutes generated by this method are more complete,have a lower error rate,and more able to ensure the quality of the meeting minutes.
作者
赵常辛
ZHAO Changxin(CHN Energy Wuhai Energy Laoshidan Colliery,Wuhai 016000,China)
出处
《电子设计工程》
2025年第24期108-112,共5页
Electronic Design Engineering
基金
国家能源集团乌海能源老石旦煤矿科技项目(TH(G)0101012207008)。
关键词
语音识别技术
会议记录报告
预处理
特征提取
双层神经网络
智能生成
speech recognition technology
meeting minutes report
pre⁃processing
feature extraction
double layer neural network
intelligent generation
