Speech deepfake detection has achieved remarkable success in clean environments but faces significant challenges in complex, real-world scenarios where speech is often mixed with background music or noise. Current state-of-the-art methods rely on semantic features from self-supervised learning (SSL) models, which often fail when processing non-speech or mixed-source audio. In this paper, we first introduce MixFake, a large-scale benchmark dataset designed to simulate diverse acoustic environments with varying SNR levels and mixed authenticity components. To address the "semantic-centric" limitation, we propose a Multi-stream Prompt Tuning framework that injects signal-level priors into SSL backbones. By integrating base, frequency, and texture streams through deep prompt injection, our model effectively captures acoustic artifacts. Experimental results demonstrate that our method significantly outperforms existing baselines, achieving a 0.95% EER in foreground detection and a substantial 7.72% absolute improvement in complex background detection tasks. Our dataset and code are available at https://github.com/saltfish233/MixFake.

翻译：语音深度伪造检测在干净环境下取得了显著成功，但在现实复杂场景中（如语音常与背景音乐或噪声混合）面临重大挑战。当前最先进的方法依赖自监督学习（SSL）模型提供的语义特征，但这类特征在处理非语音或混合源音频时往往失效。本文首先提出MixFake——一个大规模基准数据集，旨在模拟不同信噪比和混合真实性成分的多样化声学环境。为解决“语义中心”局限，我们提出多流提示调优框架，将信号级先验注入SSL主干网络。通过基础流、频率流和纹理流的深度提示注入，模型有效捕获声学伪影。实验结果表明，我们的方法显著优于现有基线，在前景检测中达到0.95%的等错误率，在复杂背景检测任务中实现7.72%的绝对提升。数据集与代码已开源：https://github.com/saltfish233/MixFake。