DirectAudioEdit: Inversion-Free Text-Guided Audio Editing via Diffusion Prediction Contrast 事件

DirectAudioEdit: Inversion-Free Text-Guided Audio Editing via Diffusion Prediction Contrast arXiv:2606.07356v1 Announce Type: cross Abstract: Text-guided audio editing aims to modify the language-specified acoustic content while preserving edit-irrelevant source components. Existing training-free methods typically rely on inversion-based editing. While inversion-free editing is appealing as it decreases computational overhead and reconstruction errors, it remains largely unexplored for audio ed