通过双分支对抗解耦，实现跨语言语音验证中的说话人与语言分离。
Dual-LoRA: Parameter-Efficient Adversarial Disentanglement for Cross-Lingual Speaker Verification
- 在冻结预训练模型中注入可训练的LoRA适配器，实现参数高效微调。
- 在TidyVoice数据集上达到0.91%的验证误错率，排名第三。
- 适合需要低资源跨语言说话人验证的研究者和工业应用。
跨语言说话人验证面临严重的语言-说话人纠缠问题，尤其在相同说话人不同语言的语句正确接受、不同说话人同语言语句错误拒绝这一极端场景下性能显著下降。标准对抗解耦会降低说话人区分能力；盲判别器会误罚仅与语言相关但对说话人识别重要的特征。为此，我们提出Dual-LoRA，将可训练的任务因子化LoRA适配器注入冻结的预训练骨干网络。核心创新是语言锚定对抗器：通过显式语言分支锚定判别器，使对抗梯度聚焦于真实语言线索而非任意相关性，从而保留关键说话人特征。在TidyVoice基准上，系统达到0.91%的验证等错误率（EER），在官方挑战赛中位列第三。
原文摘要 · Abstract (English)
Cross-lingual speaker verification suffers from severe language-speaker entanglement. This causes systematic degradation in the hardest scenario: correctly accepting utterances from the same speaker across different languages while rejecting those from different speakers sharing the same language. Standard adversarial disentanglement degrades speaker discriminability; blind discriminators inadvertently penalize speaker-discriminative traits that merely correlate with language. To address this, we propose Dual-LoRA, injecting trainable task-factorized LoRA adapters into a frozen pre-trained backbone. Our core innovation is a Language-Anchored Adversary: by grounding the discriminator with an explicit language branch, adversarial gradients target true linguistic cues rather than arbitrary correlations, preserving essential speaker characteristics. Evaluated on the TidyVoice benchmark, our system achieves a 0.91% validation EER and achieves 3rd place in the official challenge.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。