基于深度学习的ChatGPT中文语音自动识别方法  

ChatGPT Chinese speech automatic recognition method based on deep learning

作　　者：郑瑶 ZHENG Yao(Municipal and Environmental Engineering Institute,Jilin University of Architecture and Technology,Changchun 130114,China)

机构地区：[1]吉林建筑科技学院市政与环境工程学院,吉林长春130114

出　　处：《无线互联科技》2024年第17期91-93,共3页Wireless Internet Science and Technology

基　　金：吉林省高等教育学会,项目名称:ChatGPT在高质量高等教育发展中的应用研究,项目编号:JGJX2023C150。

摘　　要：针对传统中文语音信息识别对语音噪声处理不足、影响识别效果的问题,文章提出了一种基于深度学习的ChatGPT中文语音自动识别方法。该方法首先对原始语音信号进行离散傅里叶变换,并利用美尔标数将线性频率映射到美尔非线性频谱,提取中文语音频谱特征。然后,计算倒谱的频率响应和,引入深度卷积神经网络对提取的特征进行匹配分析,确定最终的识别结果。最后,应用实验证明所提方法的先进性,测试结果表明,该方法应用后,噪声对识别结果的影响被显著减弱,整体CCER均值为9.34%,应用效果较好。In view of the problem that the traditional Chinese speech information recognition handles the speech noise insufficient and affects the recognition effect,this paper proposes a ChatGPT Chinese speech automatic recognition method based on deep learning.This method first performs the discrete Fourier transform of the original speech signal,and maps the linear frequency to the linear nonlinear frequency spectrum to extract the Chinese speech spectrum features.Then,the frequency response sum of the inverted spectrum is calculated,and a deep convolutional neural network is introduced to match the extracted features to determine the final identification results.Finally,the experiment is applied to prove the advancement of the proposed method.The test results show that the influence of noise on the recognition result is significantly weakened,and the overall CCER average of the method is 9.34%,and the application effect is good.

关 键 词：深度学习 中文语音识别 离散傅里叶变换 美尔标数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]