语音伪造与鉴伪的发展与挑战

陶建华; 傅睿博; 易江燕; 王成龙; 汪涛

引用本文：

陶建华,傅睿博,易江燕,王成龙,汪涛.语音伪造与鉴伪的发展与挑战[J].信息安全学报,2020,5(2):28-38 [点击复制]
TAO Jianhua,FU Ruibo,YI Jiangyan,WANG Chenglong,Wang Tao.Development and Challenge of Speech Forgery and Detection[J].Journal of Cyber Security,2020,5(2):28-38 [点击复制]

本文已被：浏览 11054次下载 12944次	码上扫一扫！
语音伪造与鉴伪的发展与挑战
陶建华^1,2,3, 傅睿博^1,2, 易江燕¹, 王成龙¹, 汪涛^1,2
0 字体:加大+\|默认\|缩小-
(1.中国科学院自动化研究所模式识别国家重点实验室北京中国 100190;2.中国科学院大学人工智能技术学院北京中国 100190;3.中国科学院自动化研究所中国科学院脑科学与智能技术研究中心北京中国 100190)

摘要:

本文对语音伪造与鉴伪的发展进行了梳理与阐释。针对语音伪造的适用场景与关键技术点，分别对身份风格伪造、音色与韵律伪造、语音模拟三大核心语音伪造技术的基本概念、发展历程、优势与不足进行梳理与分析。针对语音伪造的应对技术语音鉴伪技术，首先介绍整理了针对性较强、面向参数式语音伪造、拼接式语音伪造与语音模拟技术框架的应对技术，在此基础上介绍了具有普适性更强的基于深度鉴别网络语音鉴伪研究进展。在此基础上，本文针对语音伪造技术所面临口语化、低资源的挑战，对未来多风格、低成本、鲁棒性发展趋势进行分析。对于语音鉴伪，本文从语料库、特征挖掘、异常检测三个角度对未来的研究重点进行诠释。

关键词: 语音伪造语音鉴伪发展与挑战

DOI：10.19363/J.cnki.cn10-1380/tn.2020.02.03

投稿时间：2019-12-31修订日期：2020-03-10

基金项目:本课题得到国家重点研发计划（No.2018YFB1005003），国家自然科学基金（No.61831022，No.61771472，No.61773379，No.61901473），cas-inria院双边合作项目资助（No.173211KYSB20190049）。

Development and Challenge of Speech Forgery and Detection

TAO Jianhua^1,2,3, FU Ruibo^1,2, YI Jiangyan¹, WANG Chenglong¹, Wang Tao^1,2

(1.National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China;2.School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100190, China;3.CAS Center for Excellence in Brain Science and Intelligence Technology, Institute of Automation, Chinese Acadeng of Sciences, Beijing 100190, China)

Abstract:

This paper reviews and explains the development ofspeech forgery and forgery detection. According to the applicable scenarios and key technology points of deep speech forgery, the basic concepts, development process, advantages and disadvantages of three core speech forgery technologies, namely identity style forgery, timbre and rhythm forgery, and speech simulation, are analyzed. In view of the response technology of speech forgery, this paper first introduces the countermeasures of speech forgery, which is source limited, and then introduces the research progress of speech forgery based on depth identification network with better generalization. To sum up, this paper analyzes the development trend of multi-style, low cost and robustness in the future, aiming at the challenge of colloquialism and low resource faced by speech forgery technology. As for speech detection, this paper interprets the future research focus from corpus, feature mining and anomaly detection.

Key words: speech forgery speech forgery detection development and challenge