面向色情音频检测的内容分类研究

Research on Content Classification for Pornographic Audio Detection

色情音频检测是互联网信息安全中的重要组成部分,为国内外色情音频检测研究者提供良好的数据支撑。论文介绍并公开了带有色情及非色情标签的1min或30s音频片段数据集,并提出了语音识别技术和文本分类技术相结合的检测方法,实现了对色情音频自动过滤或预警。实验中论文所提方法在真实音频数据集上最高可以达到97.3%的分类正确率。

Pornographic audio detection is an important part of Internet information security.In order to provide a data support for pornographic audio detection researchers.This paper introduces and discloses a 1 minute or 30 seconds audio clip dataset with pornographic and non-pornographic tags and proposes a detection method combining speech recognition technology and text classification technology,realizes automatic filtering or warning of pornographic audio.The method proposed in this paper can achieve a maximum classification accuracy of 97.3%on the database.