Deep Learning and Large Audio/Vision Language Models for Restoration of Historical Audio-Visual Archives

This project aims to develop advanced deep learning methods and large audio/vision language models for the restoration, enhancement, and preservation of historical audio-visual archives, enabling renewed access to Europe’s cultural memory. By leveraging state-of-the-art AI techniques in image, video and speech processing, and generative models, the project will address degradation issues such as noise, scratches, missing frames, low resolution, and distorted audio in legacy film, video, and sound recordings.

The proposed framework will integrate multimodal restoration pipelines capable of jointly enhancing visual and audio quality while preserving historical authenticity. Techniques such as super-resolution, inpainting, audio denoising, and cross-modal reconstruction will be combined with explainable AI to ensure transparency and fidelity to original content. Human-in-the-loop validation with archivists and cultural experts will ensure ethical and culturally sensitive restoration outcomes.

Aligned with Horizon Europe Cluster 2 priorities, the project supports cultural heritage preservation, digital transformation of the cultural and creative sectors, and wider citizen access to European history. It fosters trustworthy and responsible AI for cultural applications, enabling scalable, cost-efficient restoration of archival collections. Expected outcomes include open restoration models, interoperable tools for heritage institutions, and guidelines for ethical AI-driven preservation of audiovisual heritage.