基于深度学习的语音信号去噪方法研究  

Research on Speech Signal Denoising Methods Based on Deep Learning

作　　者：国腾飞 徐骁 陈静 郭宝军 康宪芝 GUO Tengfei;XU Xiao;CHEN Jing;GUO Baojun;KANG Xianzhi(Cangzhou Jiaotong College,Cangzhou 061199,China)

机构地区：[1]沧州交通学院,河北沧州061199

出　　处：《电声技术》2024年第6期44-46,共3页Audio Engineering

基　　金：河北省高等教育教学改革与研究项目(2021GJJG639);中国民办教育协会课题(CANFZG23102);河北省高等学校科学研究项目(QN2024168);河北省高等教育教学改革研究项目(2021GJJG640);沧州交通学院校级科研项目(CJ202301001);沧州交通学院校级教改项目(CJ202302004)。

摘　　要：为提升语音去噪技术的整体性能,提出基于深度学习的语音信号去噪方法。通过引入Adam优化长短期记忆(Long Short-Term Memory,LSTM)网络,提高了语音去噪的效率和效果。实验部分使用TIMIT数据集进行验证,并对比了传统LSTM模型和优化后的LSTM模型的性能。实验结果表明,优化后的LSTM模型的信噪比(Signal-to-Noise Ratio,SNR)和均方误差(Mean Squared Error,MSE)均优于传统模型。To improve the overall performance of speech denoising technology,a deep learning based speech signal denoising method is proposed.By introducing Adam to optimize Long Short-Term Memory(LSTM)networks,the efficiency and effectiveness of speech denoising have been improved.The experimental part was validated using the TIMIT dataset,and the performance of the traditional LSTM model and the optimized LSTM model were compared.The experimental results show that the optimized LSTM model has better Signal-to-Noise Ratio(SNR)and Mean Squared Error(MSE)than traditional models.

关 键 词：深度学习 长短期记忆(LSTM) 语音信号 去噪 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]