오디오 딥페이크 탐지를 위한 안전한 워터마킹
Secure Watermarking for Audio Deepfake Detection
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
디지털 기술의 발전으로 딥페이크 기법은 변조된 음성을 생성하여 디지털 통신의 무결성에 심각한 위험을 초래하고 있다. 이러한 문제를 해결하기 위해 본 논문에서는 암호학적으로 안전한 워터마크를 오디오 파일에 직접 삽입하는 디지털 오디오 워터마킹 기법을 제안한다. 본 접근법은 암호화 해시 함수를 활용하여 비인가된 수정에 극도로 민감한 고유 워터마크를 생성한다. 이 워터마크는 딥페이크 조작이 발생할시, 즉시 파괴되어 변조의 명확한 표시를 제공한다. 신호 대비 잡음비 (Signal-to-Noise Ratio)과 원본과 워터마킹된 오디오의 상호 상관 (Cross Correlation) 평가 지표를 사용하여 워터마킹 시스템의 성능을 평가한 결과, 본 워터마크 방식은 높은 수준의 오디오 무결성을 보장한다. 또한 StarGAN 음성 변환, PPG 음성 변환 등 실제 딥페이크 공격에 대해 테스트를 거친 결과, 제안기법으로 워터마킹이 삽입된 음성은 딥페이크 변조 후 즉시 파괴됨을 검증했다. 본 연구 결과는 오디오의 무결성과 딥페이크 저항성을 크게 필요로 하는 미디어, 법률, 안전한 통신 등 다양한 분야에 적용 가능할 것으로 기대된다.
Abstract
With the advancement of digital technology, deepfake techniques have increasingly posed serious threats to the integrity of digital communication by generating manipulated voices. To address this issue, we propose a cryptographically secure digital audio watermarking technique that directly embeds a watermark into audio files. Our approach utilizes cryptographic hash functions to create a unique watermark that is highly sensitive to unauthorized modifications. This watermark is destroyed immediately when a deepfake manipulation occurs, providing a clear indication of tampering. We evaluate the performance of the watermarking system using signal-to-noise ratio and cross-correlation metrics to measure the similarity between the original and watermarked audio. The results show that our watermarking method ensures a high level of audio integrity. Furthermore, tests against real-world deepfake attacks, such as StarGAN voice conversion and PPG voice conversion demonstrate that the watermarked audio is immediately destroyed after deepfake manipulation. We expect that the proposed watermarking scheme is applicable across various fields requiring strong audio integrity and resistance to deepfake tampering, including media, legal sectors, and secure communications.
참고문헌 (17)
- Masood, Momina, "Deepfakes generation and detection: State-ofthe-art, open challenges, countermeasures, and way forward", Applied intelligence, 53(4), pp.3974~4026 (2023)
- Dixit, Abhishek, "Review of audio deepfake detection techniques: Issues and prospects", Expert Systems, 40(8), pp.e13322~ (2023)
- Mahmoud, Khaled W, "Frequency domain watermarking : An overview", Int. Arab J. Inf. Technol, 2(1), pp.33~47 (2005)
- Liu, Zhenghui, "Patchwork-based audio watermarking robust against de-synchronization and recapturing attacks", IEEE transactions on information forensics and security, 14(5), pp.1171~1180 (2018)
- Hu, Hwai-Tsu, "Robust, transparent and highcapacity audio watermarking in DCT domain", Signal Processing, 109, pp.226~235 (2015)
- Saadi, Slami, "Novel secured scheme for blind audio", speech norm-space watermarking by Arnold algorithm, 2019, pp.154, 74~86
- Pavlović, Kosta, "Speech watermarking using deep neural netw", 2020 28th Telecommunications Forum (TELFOR), pp.1~4 (2020), IEEE
- Kaneko, Takuhiro, "Starganvc2:Rethinking conditional methods for stargan-based voice conversion", Interspeech 2019 (2019)
- Li, Zhonghao, "Ppg-based singing voice conversion with adversarial representation learning", ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.7073~7077 (2021), IEEE
- Yoshida, Hirotaka, "Analysis of a SHA-256variant", International Workshop on Selected Areas in Cryptography, pp.245~260 (2005), Springer Berlin Heidelberg
- Mascagni, Michael, "Algorithm 806: SPRNG:A scalable library for pseudorandom number generation", ACM Transactions on Mathematical Software (TOMS), 26(3), pp.436~461 (2000)
- Kowalczuk, Yann, "Evaluation of digital watermarking on subjective speech quality", Scientific Reports (2021)
- Vorburger, Theodore V., "Applications of cross-correlation functions", Wear, 271, pp.529~533 (2011)
- Rai, Anil, "Analysis of three pitch-shifting algorithms for different musical instruments", 2019 IEEE Long Island Systems, Applications and Technology Conference (LISAT) (2019), IEEE
- Panayotov, Vassil, "Librispeech:an asr corpus based on public domain audio books", 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP), pp.5206~5201 (2015), IEEE
- Zhang, Yan, "Selected features for classifying environmental audio data with random forest", The Open Automation and Control Systems Journal, 7(1), pp.135~142 (2015)