Overview
PSMs are facing the need to raise the quantity and the quality of content they create to address linear and non-linear channels. To do so, they are adopting agile technologies to accelerate the production and the management of massive amounts of data. The three pillars of these evolutions are data management, computing architectures, and machine learning. EBU groups are addressing these three pillars by providing:
- Development of open source tools;
- Development of good practices and standards;
- Dissemination.
An overview of the Group's activities and events in just a few slides: AIM overview
Groups
The umbrella group that oversees the Metadata and AI domain works, maximizing knowledge sharing and collaboration. This group meets monthly.
The EBU and several Members have created and continue to develop a tool for benchmarking machine learning-based applications. After designing and publishing a tool to benchmark Speech-To-Text (STT) systems, a dataset to evaluate face recognition application. and a face recognition system for TV conetnt, the group is now working on LLM, RAG and agents.
The main activity of this group is to support Members in the domain of metadata, which plays an essential role in media. To do so, the group develops specifications and promotes innovation, such as the use of semantic technologies.
AI and Automatic Metadata Extraction (suspended)
The group has been suspended, and the presentations are still available in the workspace. Work addressed in this group includes metadata schemas, the capabilities and performance of automatic metadata extraction tools, and the development of machine learning algorithms and related tools.
Media Cloud Microservice Architecture (MCMA)
MCMA is a serverless/micro-services strategy for media. It aims to help developers to move to "Function as a Service" architectures. Standardization of MCMA has started with SMPTE. This project provides open-source libraries.
Communities
Currently, the following deliverables are planned (green indicates the deliverable has already been delivered). Note that deliverables are dependent on enough participation in the work and that the planning is subject to change. New deliverables are added regularly.
2024
-
Knowledge sharing on metadata/data/AI technology -
DataTech Seminar 2024 -
Development and deployment of the meta-radio application on the AI HUB
-
AI Benchmarking Group: experiment on Q&A for proprietary content with LLMs, RAGs, agents - write report.
-
SMPTE task force on AI
-
Development of the cloud-hosted AI-HUB to evaluate, expose, and exchange AI applications
-
Development and deployment of a face recognition system for TV programme on the AI HUB
-
SMPTE task force on AI and metadata
-
Update the EBUCorePlus
-
Research projects with the EPFL
-
Training course for the EBU Academey
-
Specify the Metadata model or the EU project TEMS
-
Development of AI models for analysing editorial content for Vera.ia
-
Develop a demo for IBC
2023
-
Development of datasets to evaluate facial recognition systems: the biggest for AV content!
-
Publication of the first release of the EBUCorePlus : the EBU ontology for media
-
DataTech Seminar 2023
-
ETC/SMPTE task force on AI: engineering report AI for Media
-
Development of the cloud-hosted AIM platform to evaluate, expose, and exchange AI applications
-
Development of an open-source Facial Recognition Framework for Video
-
Knowledge sharing on metadata/data/AI technology
-
Maintenance of the MCMA libraries and MAM
-
Advanced Studies on Machine Learning for Members
2022
-
Open source demo MAM based on serverless MCMA framework (Q2 2022)
-
MDN Workshop 2022 ( Q2 2022)
-
Fake News detector for English text: API open to Members (Q2 2022)
-
EBUCore+ Demonstrator Kit: a cloud-hosted demo (Q3 2022)
-
Cloud-hosted Metadata Exchange Platform for archive (Q3 2022)
-
Knowledge Sharing on MetaData and AI (Q4 2022)
-
Organisation of DataTech Seminar 2023
-
Development of an open-source Facial Recognition Framework for Video (Q1 2023)