arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2510.03387cs.SDeess.AS
收藏

合成音频取证评估(SAFE)挑战

Synthetic Audio Forensics Evaluation (SAFE) Challenge

STR Woburn, MA(STR沃本麻省);Aptima, Inc.(Aptima公司);Drexel University(德雷塞尔大学);ULRI Digital Safety Research Institute(ULRI数字安全研究所)

Kirill Trapeznikov, Paul Cummer, Pranay Pherwani, Jai Aslam, Michael S. Davinroy, Peter Bautista, Laura Cassani, Matthew Stamm, Jill Crisman

AI总结:

针对合成语音检测难题,提出SAFE盲测框架,含90小时音频、21个来源、17个TTS模型和3个任务,评估模型在原始、处理和清洗场景下的表现。

摘要:

先进文本转语音(TTS)模型生成的合成语音日益逼真,加之后期处理与清洗技术,对音频取证检测构成了重大挑战。本文介绍了SAFE(合成音频取证评估)挑战,这是一个完全盲测的评估框架,旨在跨渐进困难场景对检测模型进行基准测试:原始合成语音、处理后的音频(如压缩、重采样)以及旨在逃避取证分析的清洗音频。SAFE挑战共包含90小时音频和21,000个音频样本,分布在21个不同真实来源、17个不同TTS模型和3个任务中。我们展示了挑战、评估设计与任务、数据集细节,以及对当前方法优势与局限的初步见解,为推进合成音频检测研究奠定了基础。更多信息请访问https://stresearch.github.io/SAFE/。

英文摘要:

The increasing realism of synthetic speech generated by advanced text-to-speech (TTS) models, coupled with post-processing and laundering techniques, presents a significant challenge for audio forensic detection. In this paper, we introduce the SAFE (Synthetic Audio Forensics Evaluation) Challenge, a fully blind evaluation framework designed to benchmark detection models across progressively harder scenarios: raw synthetic speech, processed audio (e.g., compression, resampling), and laundered audio intended to evade forensic analysis. The SAFE challenge consisted of a total of 90 hours of audio and 21,000 audio samples split across 21 different real sources and 17 different TTS models and 3 tasks. We present the challenge, evaluation design and tasks, dataset details, and initial insights into the strengths and limitations of current approaches, offering a foundation for advancing synthetic audio detection research. More information is available at \href{https://stresearch.github.io/SAFE/}{https://stresearch.github.io/SAFE/}.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑