The European Broadcasting Union EBU’s annual Metadata Network MDN Workshop 2019 was organised in June 2019 in Geneva. It gathered 88 experts from different media companies, research institutes and vendor companies to share latest experiences and ideas related to metadata in the media business.
Thematically related to MeMAD, many broadcasters are already working on automatic content analysis. They presented their interesting recent projects in developing and using machine learning techniques to automatically analyse their media content to achieve business benefits such as increased findability of archive content.
For example, Deutsche Welle presented an semi-automatic translation system for news videos. The system uses automatic speech recognition to create a transcript of what is said and then the transcript is automatically translated to the requested language. The human in the loop is then checking the language – only incorrect claims or major language mistakes are fixed, stylistic details are left as they are.
Many broadcasters presented their current tests in using automatic content analysis to assist e.g. archive search, metadata creation and increasing the end-user experience. For example, the Swiss public broadcaster RTS is developing and already using in production automatic content analysis technologies to improve the findability of their audiovisual archive. The system indexes the archive content for example with the help of face recognition, visual feature extraction and speech recognition to make the content more accessible.
The Finnish Broadcasting Company Yle (one of the partners of the MeMAD project) presented their findings from the “Metadata machine” project. The goal of the project was to identify business cases for currently commercially available automatic content analysis technologies, such as various speech recognition services, visual feature recognition, optical character recognition, sound recognition and others. As a result, tens of use cases were identified and tested with existing technologies, which were reported in the MDN workshop.
The Italian public broadcaster RAI has been developing a platform for using multiple commercial content recognition services. One of the goals of this work is to provide a framework for testing and comparing multiple service providers, so that the most suitable providers can be found. Also EBU (in co-operation with e.g. the British BBC) is currently working on evaluation frameworks for automatic content analysis – where the first step has been tools for evaluating speech recognition services.
To summarise, many broadcasters are working on automatic content analysis. The majority of the activities are still in development phase, but first implementations are already in production, such as the Swiss broadcaster’s archive indexing system.
The MeMAD project was also present in the seminar. Simon Debacq from Limecraft demonstrated the first implementation of the MeMAD prototype by showing the results of the different services. Ismail Harrando from EURECOM presented how metadata is modelled in the MeMAD project as a knowledge graph. The former got lots of interest among the audience.
Ismail Harrando from EURECOM on stage presenting MeMAD’s knowledge graph.
Read also: Workshop grows in line with role of data and AI, EBU website.
Main photo credit: Guilhem Vellut Creative commons Attribution 2.0 Generic (CC BY 2.0).