Investigation on secure speaker verification based on vocal liveness detection

2016 Fiscal Year Final Research Report

• Project/Area Number: 26280066
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Perceptual information processing
• Research Institution: National Institute of Informatics
• Principal Investigator: Yamagishi Junichi 国立情報学研究所, コンテンツ科学研究系, 准教授 (70709352)
• Co-Investigator(Kenkyū-buntansha): 松井 知子 統計数理研究所, モデリング研究系, 教授 (10370090) ; 越前 功 国立情報学研究所, コンテンツ科学研究系, 教授 (30462188) ; 小野 順貴 国立情報学研究所, 情報プリンシプル研究系, 准教授 (80334259) ; 塩田 さやか 首都大学東京, システムデザイン研究科, 助教 (90705039)
• Project Period (FY): 2014-04-01 – 2017-03-31
• Keywords: 生体認証 / 話者照合 / 詐称 / 生体検知 / 音声合成 / 声質変換

Outline of Final Research Achievements

Among various biometrics, speech-based biometrics called speaker verification becomes popular recently and has been adopted for various applications in society and industry because it can be conveniently utilized based on a microphone only. However, it is also reported that spoofing of speaker verification is possible using recently advanced speech synthesis and voice transformation techniques. For this problem, we have introduced a novel technique called liveness detection or anti-spoofing for speaker verification systems to verify if an inputted speech signal is really produced by a live human or not. We have also constructed the first large-scale spoofed speech database in the world and have organized events where participants compete the accuracy of their anti-spoofing techniques on the same corpus. Moreover, we showed that it is possible to discriminate live human speech from speech replayed from loudspeakers using a pop noise, which is a distorted signal caused by breath.

Free Research Field

音声情報処理