IBM Watson Media hopes to launch a new AI-driven search and discovery product within the next few months to help broadcasters better exploit their archives.
David Kulczar, Senior Product Manager at IBM, revealed the plans at IBC, following the launch of a number of new solutions exploiting its AI-driven ‘video enrichment’ technology to offer benefits such as automated closed captioning, better viewer recommendations and automated highlight identification and clipping.
“One of the key issues we’re hearing from our clients is around large archives and how to better manage [them],” said Kulczar. “The next key product that we’re looking to launch is a search and discovery solution for content.”
Kulczar notes that IBM is “already starting to provide tools that allow you to have all the information you need to quickly find relevant data.” So the first step would be for Watson to scan a broadcaster’s entire archive and, using its ability to extract semantic meaning from video and related metadata, ‘learn’ what it contains and automatically tag every video clip.
The idea is then to build on that existing capability: the example Kulczar cites is a producer seeking to create a content package around Hurricane Florence, where, for instance, there is a need to find other similar hurricane footage from the past. “[The system] would quickly bring back from across your entire archive things that might be relevant.”
The next step IBM is looking at is “how do we get to the next level – maybe even helping [the producer] to build a rough cut?” Kulczar said this was “something we’re working on so it’s hard to be accurate – but it could be in the next few months.”
IBM demonstrated over the summer how Watson could be used to streamline Fox Sports’ production workflows for the 2018 FIFA World Cup, by quickly classifying, editing and accessing match highlights in near real-time – so the foundations for this type of AI intervention in the creative process arguably already exist within IBM.
Just before IBC, IBM also announced a partnership with Iris.TV, a cloud-based personalised video programming system, to launch a new AI-driven video recommendation engine. This also exploits Watson’s ability to analyse a client’s video library to extract ‘rich metadata’ or semantic meaning from audio and video content, in this case combining it with customer viewing data to provide improved recommendations.
Watson is also already capable of analysing video sound-tracks to automate the captioning process through its Watson Captioning product, providing what is in effect a live transcript.
This is capable of exceeding the accuracy of ‘human’ live transcription provided the AI is ‘trained’ ahead of time, achieving accuracy rates of between 92-96% on average.
“Out of the box it’s pretty accurate,” notes Kulczar. “But what we’ve done is we’ve tried to codify and build a lot of tooling around how you train [the AI in a particular] environment. With that it can be highly accurate.”
In the case of captioning, “the way we generally work with broadcast TV is if you have information about what’s going to be said – generally the [broadcaster] is going to have story information that’s going to be read out over the air – we can load that story into the system ahead of time so we know what’s going to be said.” Words that aren’t in Watson’s ‘dictionary’, for instance, can then be spotted and added ahead of time.