发表机构
The Chinese University of Hong Kong, Shenzhen; Amphion Technology Co., Ltd.; Huawei Technologies Co., Ltd.（香港中文大学（深圳）; Amphion科技有限公司; 华为技术有限公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
GenTraceBench基准系统评估TTS模型适配后音频深度伪造指纹的稳定性，发现DPO/GRPO保留指纹而部分SFT导致漂移，多镜头注册可显著降低验证错误率。
AI 中文摘要
现代文本转语音（TTS）系统很少以未更改的预训练模型形式部署。它们通常通过监督微调（SFT）或偏好优化（如DPO和GRPO）进行适配。这为音频深度伪造取证提出了一个实际问题：从基础生成器学到的指纹在适配后是否仍然有效？我们提出了GenTraceBench，一个受控基准，涵盖五种TTS架构、16种预训练/后训练变体，以及使用固定文本和说话人提示生成的49,728条话语。在“基于基础训练、基于适配测试”协议下，我们评估了二元检测、闭集归因和开集验证。DPO和GRPO通常保留指纹，而某些SFT和预训练数据变化会导致显著漂移；效应大小在三种取证骨干网络上有所不同。重复训练运行确认了最大的W2V-BERT归因下降，而一个具有可比语音质量的数据混合对照表明，组成变化不一定会导致漂移。在W2V-BERT验证中，多镜头注册将SFT条件下的等错误率（EER）从44.4%降至11.0%，而仅SingNet条件仍保持45%或更高的EER。
英文摘要
Modern text-to-speech (TTS) systems are rarely deployed as unchanged pre-trained models. They are often adapted through supervised fine-tuning (SFT) or preference optimization such as DPO and GRPO. This raises a practical question for audio deepfake forensics: do fingerprints learned from a foundation generator remain valid after adaptation? We present GenTraceBench, a controlled benchmark spanning five TTS architectures, 16 pre-/post-training variants, and 49,728 utterances generated with fixed texts and speaker prompts. Under a train-on-foundation, test-on-adapted protocol, we evaluate binary detection, closed-set attribution, and open-set verification. DPO and GRPO generally preserve fingerprints, whereas some SFT and pre-training-data changes cause substantial drift; effect sizes vary across three forensic backbones. Repeated training runs confirm the largest W2V-BERT attribution drop, while a data-mixture control with comparable speech quality shows that composition change need not cause drift. In W2V-BERT verification, multi-shot enrollment reduces EER for the SFT condition from 44.4% to 11.0%, whereas the SingNet-only condition remains at or above 45% EER.
Comments5 pages, 2 figures, 4 tables. Accepted to the 15th International Symposium on Chinese Spoken Language Processing (ISCSLP 2026)
