Introduction 

Recent advances in generative models have significantly improved the realism of synthetic audio, making audio deepfake detection an increasingly critical research problem. Although substantial progress has been made in speech deepfake detection, a notable gap remains between current research settings and real-world deployment scenarios. In practice, audio may be captured by diverse devices (e.g., mobile phones, wearable devices, and in-vehicle systems), transmitted through different channels, and affected by varying acoustic conditions, resulting in significant domain shifts. Meanwhile, emerging generation paradigms, particularly audio large language models and neural codec-based synthesis frameworks, are rapidly expanding the diversity of synthetic audio. As a result, countermeasures trained on existing public datasets often struggle to generalize to unseen generation methods and real-world conditions, highlighting the need for robust speech deepfake countermeasures. 

Beyond speech, another limitation of current research is its strong reliance on audio modality-specific frameworks. Most studies focus on individual audio types, and design dedicated detectors for each modality. However, real-world audio can belong to any type, including speech, sound, singing voice, or music. Deploying separate detectors for different audio types is often impractical and fails to provide a unified solution for real-world authenticity verification. Therefore, there is a growing need for universal audio deepfake countermeasures that can generalize across diverse audio types and synthesis mechanisms within a single framework. 

To address these challenges, we introduce the AT-ADD Grand Challenge on All-Type Audio Deepfake Detection, which aims to promote the development of robust speech countermeasures and universal audio deepfake detection methods.