多场景下中文录音回放语音数据集的设计与分析  

Design and Analysis of A Chinese Replay Spoofing Attack Dataset in Multiple Scenes

作　　者：周颖慧[1] 刘亚丽[1] ZHOU Ying-hui;LIU Ya-li(Communication Acoustic Laboratory,Communication University of China,Beijing 100024,China)

机构地区：[1]中国传媒大学传播声学研究所

出　　处：《中国传媒大学学报（自然科学版）》2020年第1期30-35,共6页Journal of Communication University of China：Science and Technology

基　　金：中央高校基本科研业务费专项资金(CUC2019B030)

摘　　要：录音回放语音数据集是说话人识别下录音回放检测技术的数据基础与研究对象,随着录音回放检测技术的发展,有关标准数据集的需求也愈加迫切。本文以说话人识别下防录音回放攻击为应用背景,根据录音回放检测技术的相关需求,建立了一套多设备、多环境、多语料的针对中文的录音回放语音数据集,并通过听音实验与声学参数的初步分析,对数据集质量进行综合评价,为进一步建立更为完备的录音回放数据集提供一定的参考。The replay spoofing attack datasets are the basis of the playback detection technology.With the development of the playback detection technology,the demand for standard datasets is becoming more and more important.In this paper,the existing replay spoofing attack datasets are analyzed and compared.According to the relevant requirements of the playback detction technology,a set of replay spoofing attack dataset for Chinese,multi-device,multi-environment and multi-linguistic materials is established,and related subjective experiments are done.With the objective analysis,the overall evaluation of the dataset provides a data foundation for the better development of the playback detection technology.

关 键 词：录音回放语音数据集 说话人识别 听音实验 声学参数分析 

分 类 号：TN912.34[电子电信—通信与信息系统]