AI and Automatic Metadata Extraction

AI and Automatic Metadata Extraction (AME) in Production.

More than ever, metadata is indispensable for all processes from production to distribution.

As we move to (micro)service-based production, broadcasters need to consider the widespread adoption of automatic information extraction tools, incl. in the cloud, as new processes in agile workflows.

These tools can be used to produce more information (including structured metadata) that is needed by modern production systems, at a lower cost.

AME is part of the EBU "Metadata and Artficial Intelligence" activities. AME tools are characterised by the data they can extract.

Ideally, tool capabilities should be registered to facilitate service registration and discoverability in microservice-based architectures (for more Information, please visit the page of the EBU "Media Cloud and Microservices Architecture" project) .

Artificial intelligence techniques such as machine learning and deep learning or neuronal networks are behind most AME tools.

As an example, the EBU "AI Benchmarking" project develop tools to evaluate tools that do speech-to-text transcription and entity recognition.

The EBU "AI Data Pool" project proposes a framework to share resources for training and assessing AI tools.

AME_ow_logo1-1.jpg

The topic of Automatic Metadata Extraction is part of the EBU's Strategic Programme on Production.

Main activities:

  • Help Members move towards the adoption of AI-based automatic information extraction tools.
  • Identify and evaluate/benchmark tools investigating how they can help with e.g. archive management applications, multi-purpose / multi-channel productions and news production.
  • Update the AME cards (https://github.com/ebu/ame-cards) through which tools capabilities should be exposed.
  • Update EBU members' use cases.
  • Pool test and ground truth material.

Related topics