Partially Edited Media:
Perspectives from Synthesis and Defense

Recent advances in generative AI have enabled partially edited media content—text, speech, images, video, and multimodal content in which only selected regions are modified while the rest of the original signal remains intact. This emerging paradigm is fundamentally different from full synthesis and creates urgent research needs in both generation and defense. On the generation side, key challenges include seamless consistency between edited and untouched regions, controllable localized editing, cross-modal synchronization, and unified evaluation across media types. On the defense side, partially edited media is harder to detect than fully synthetic content because authentic and manipulated regions coexist within the same sample, requiring new methods for detection, localization, attribution, provenance analysis, and edit reconstruction. A dedicated special session will provide a timely forum to bring together researchers from signal processing, machine learning, multimedia analysis, forensics, and AI safety to define this new problem space.

The committed papers in this session represent a strong and diverse body of work spanning speech editing, image and video manipulation, text editing, and forensic analysis. Collectively, they address both foundational generation problems and critical security implications, with potential impact on next-generation media editing tools, trustworthy content authentication, and robust AI safety systems. This cross-domain scope and dual focus on creation and defense make the session highly relevant and impactful for the ICASSP community.

The most relevant IEEE Signal Processing Society Technical Committees are Machine Learning for Signal Processing (MLSP), Audio and Acoustic Signal Processing (AASP), Speech and Language Processing (SLP), Image, Video, and Multidimensional Signal Processing (IVMSP), Multimedia Signal Processing (MSP), and Information Forensics and Security (IFS). The proposed session sits at the intersection of these areas.

These papers are better suited for a special session than for a regular session because the topic is emerging, interdisciplinary, and not yet represented as a regular session in ICASSP. In a regular session, related papers on partially edited media would likely be scattered across modality-specific tracks, even when they share common model ideas such as diffusion-based generation or consistency constraints. Likewise, generation and detection are tightly coupled: improvements in editing often introduce new forensic challenges. A special session can bring these dispersed but closely related efforts together in one focused forum, enabling cross-modal comparison and community building, outcomes that are difficult to achieve through isolated regular papers.

Submission Link: please fill this form ASAP to get detailed Instruction and submission link.

Instruction: We are following the same Author Instructions for IEEE ICASSP 2027.

NOTE (updated on Sep. 20 AoE)

Please note that ONLY a limited number of papers can be submitted to the special session at ICASSP, and we have now reached the submission limit.

If you didn't get the specific submission for the special session by Sep. 20, please submit your paper to the Regular Session instead, and fill out this form to join the waitlist of the special session. If any spots become available, we may move your paper from the regular session to the special session.

Papers in the special session and regular session will undergo a similar review process.

Thank you, and best wishes for your submission!