改进遗传算法的电子会议汉语语音识别方法  

Improved genetic algorithm for Chinese speech recognition method in electronic conference

作　　者：杨艺西 武志栋 袁洲 陈思平 何宇泽 YANG Yixi;WU Zhidong;YUAN Zhou;CHEN Siping;HE Yuze(Information and Communication Branch of State Grid Corporation of China,Beijing 100000,China)

机构地区：[1]国家电网有限公司信息通信分公司,北京100000

出　　处：《电子设计工程》2024年第18期132-135,140,共5页Electronic Design Engineering

基　　金：国家电网有限公司信息通信分公司科技项目(536826210012)。

摘　　要：语音识别普遍存在识别不准确、不全面的问题,影响电子会议汉语记录的质量。面对这种情况,为提高语音识别性能,提出一种改进遗传算法的电子会议汉语语音识别方法。该方法通过预加重、分帧加窗以及去噪,预处理语音信号。利用改进遗传算法选取最优语音特征,语音特征包括梅尔频率倒谱系数、短时平均能量以及频谱均值。以三个特征对应数值的标准化数值为输入,利用构建的基于改进神经网络的识别模型将语音转换为对应的汉语文字,实现语音识别。结果表明,在基于改进遗传算法的识别方法应用下,误识率最高仅为2.122%,识全率最低为95.621%,由此说明所研究识别方法的识别更为准确和全面,识别效果更好。There are many problems in speech recognition,such as inaccuracy and incomprehensiveness,which affect the quality of Chinese recording of electronic conference.In order to improve the performance of speech recognition,an improved genetic algorithm is proposed for Chinese speech recognition method in electronic conference.The method preprocesses speech signals by means of preweighting,frame-dividing windowing and denoising.The improved genetic algorithm was used to select the optimal speech features,including the Meir frequency cepstrum coefficient,short-term mean energy and spectrum mean.With the standardized values of the corresponding values of the three features as input,the recognition model based on the improved neural network is used to convert the speech into the corresponding Chinese characters to realize the speech recognition.The results show that under the application of the improved genetic algorithm,the highest error rate is only 2.122%,and the lowest accuracy is 95.621%,which indicates that the proposed recognition method is more accurate,more comprehensive and has better recognition effect.

关 键 词：改进遗传算法 电子会议 特征选取 汉语语音识别 

分 类 号：TN514.22[电子电信]