AI Algorithms for Media Cataloguing in ATRESMEDIA GROUP
by Eugenio López de Quintana, ATRESMEDIA
The constant increase in the volume of media received in television archives, and the growing demand for granularity in image searches, makes essential the use of automatic cataloging technologies.
This project includes the use of Artificial Intelligence Algorithms in the ATRESMEDIA Archive but integrated into the MAM-DAM System and fully incorporated into the daily workflows and operations of cataloguing and searching.
The scope covers the following types of content :
- Video materials based on statements by two or more consecutive speakers from
- Raw camera recordings or parts of a program.
- Complete programs that have a subtitle file.
- Photographs with predominant appearance of people.
More than 16.000 hours of video materiales and 200.000 stills have been already processed
The general approach is not exhaustive but progressive according to the real possibilities offered by the algorithms.
However, the project is not just a mere technological renovation. The objective of ATRESMEDIA with the use of AI is to transform the professional profile of the television archivists who work in the Group.
This will involve progressively phasing out manual work for some tasks as segmentation and description of people’s statements, on one side, besides the people recognition.
The hope is to redirect these resources towards new activities with the potential to add greater value, such as generating fresh content from archive materials, creating virtual universes of knowledge through ontologies that allow navigation in queries, and anticipating the information needs of users based on the news. In short, it is hoped to move from processing content to generating content and information.
Algorithms implemented:
- Automatic speech recognition
- Automatic text punctuation
- Speaker segmentation
- Content segmentation
- Voice activity detection and signal-to-noise ratio in voice segments
- Automatic subtitle resynchronisation
- Face detection and recognition
This project has recently received the 2021 Quality and Innovation Award from the Spanish Society for Scientific Information and Documentation, SEDIC, in its XV edition.
