计算机与数字工程2025，Vol.53Issue(4)：1063-1069,7.DOI:10.3969/j.issn.1672-9722.2025.04.025
基于改进NISQA的管制语音质量评价方法研究
Research on Controlled Speech Quality Assessment Method Based on Improved NISQA
摘要
Abstract
Based on the problems of gradient disappearance and low precision of feature extraction in the convolutional neural network of the NISQA model,a NISQA model based on a three-layer structure optimized convolutional neural network is proposed for regulatory speech quality evaluation.The SMU activation function is used to improve the convolutional layer to alleviate the prob-lem of gradient disappearance,the intermediate pooling method is used to improve the pooling layer to reduce feature extraction er-rors,the global pooling layer is used instead of the fully connected layer to reduce model complexity,and comparative experiments are conducted with the original NISQA model and the P.563 model proposed by ITU.Experimental results show that the proposed method improves the correlation coefficient and reduces the mean square error compared to the comparative models on the regulatory speech dataset.关键词
NISQA/卷积神经网络/语音质量Key words
NISQA/convolutional neural network/speech quality分类
信息技术与安全科学引用本文复制引用
傅强,李贵民,吴岳洲..基于改进NISQA的管制语音质量评价方法研究[J].计算机与数字工程,2025,53(4):1063-1069,7.基金项目
国家重点研发计划(编号:2021YFF0603904) （编号:2021YFF0603904）
中央高校基本科研业务费基金项目(编号:ZJ2022-004) （编号:ZJ2022-004）
中国民用航空飞行学院面上项目(编号:JG2022-06)资助. （编号:JG2022-06）
