XAI-Grounded Explanation Generation for Speech Deepfake Detection with Training-Free Multimodal Large Language Models

The article presents a training-free framework for generating grounded explanations in speech deepfake detection (SDD) by integrating explainable AI (XAI) evidence with multimodal large language models (LLMs). Utilizing the PartialSpoof dataset, the proposed method significantly enhances explanation accuracy by over 45% compared to traditional approaches, as confirmed by human evaluations and faithfulness checks. This advancement is crucial for practitioners as it improves the interpretability and trustworthiness of SDD systems, facilitating better decision-making in AI applications.