语音识别技术在会商系统中的应用研究

打开文本图片集

关键词：语音识别；卷积神经网络；会商系统中图分类号：TP391 文献标志码：ADOI:10.19968/j.cnki.hnkj.1003-5168.2025.14.008文章编号:1003-5168(2025)14-0041-04

Application Research of Voice Recognition Technologyin Conference Systems

LI Jiao （Henan Provincial Meteorological Observation Data Center,Zhengzhou 45ooo3,China）

Abstract: [Purposes] To achieve the real-time subtitle function in the meteorological consultation system with a certain recognition accuracy rate,we propose applying convolutional neural network-based voice recognition technology to the conference systems.[Methods] The article conducts an in-depth research on the convolutional neural network voice recognition technology.It adopts the method of dataset distribution processing and feature analysis to design a real-time subtitle implementation process of voice set. The system architecture includes MCU,SMC, ASR server,and recording server, forming a voice recognition platform for the conference system.Next,the Mel-Scale Frequency Cepstral Coeficients (MFCC)are used to describe and extract features from the voice signal.Finally,without compromising the recognition accuracy of the convolutional neural network,the nonlinear Sigmoid function is introduced to optimize the convolutional neural network's convergence speed.Additionally,an end-to-end Connectionist Temporal Clasification (CTC) voice algorithm optimization model isapplied to improve the accuracy of input voice recognition.[Findings] In the conference system,this trained model can accurately recognize the audio entered by the terminal input in the practical applications,achieving a recognition accuracy rate of up to 90% .[Conclusions] When applied to the conference system, this voice recognition algorithm model demonstrates excellent recognition speed and quality, providing strong support for intelligent meteorological business services.

Keywords: voice recognition; convolutional neural network; conference system

0 引言

语音识别技术的出现使得设备能够识别并理解人类的语音指令[。（剩余6368字）