Media Asset Management METADATA


Man vs machine


Can automated metadata tools help producers realise more of the value of their programming, asks Adrian Pennington


O


ne of the problems with talking about and collecting metadata is that it can mean different


things to different people at different stages in a programme’s lifecycle. A fi nished asset could, for example, contain broadly three layers of data: technical, descriptive and contextual. The fi rst, technical (frame rate,


frame size, commercial break, bit rate), is a requirement to deliver a quality video service regardless of the delivery channel; while the second, descriptive (summary, actors, director, reviews), is important for search and recommendation. According to Chuck Parker, chair-


man of the US-based Second Screen Society, the third layer – contextual (what objects are in the scene, what is happening) – is “critical for any higher-value experiences” – such as e-commerce, contextual and second- screen advertising, and discovery of new content. “It’s the most valuable and the hardest to collate,” he adds. Automated content-recognition


technologies like audio watermarking can already trigger second-screen applications by identifying that the user is watching, say, Friends, but next-generation appli- cations will be able to recognise exactly which scene you are watching. On the one


hand, this means playalong second- screen games, like The Walking Dead’s Kill Count iPhone/ iPad app, or more purely com- mercial options.


4 | Broadcast | 6 July 2012 “This information can be used to


push additional targeted advertise- ments, such as ‘Do you want to buy the dress that Rachel wears at this moment’,” says Andy Nobbs, creative content offi cer at Civolution. At issue is how much of this meta-


data should be captured by content creators. According to Joe Trainor, managing director of Deluxe Media Technologies (DMT), the earlier the metadata is extracted from an asset, the better. “The whole lifecycle of content creation is valuable metadata from original conceptual idea, story- board, scripting and into the shoot.” Producers of high-shoot-ratio pro-


gramming, such as reality shows and ob-docs, have made the logging of fi le-based clips a near essential part of close-to-transmission edits. But any further demands to capture data that is not integral to production have knock-on implications. “Simplicity is the most important


thing for producers,” says Lewis Kirk- aldie, product manager at Cinegy. “If a production is mandated to capture various complicated data, such as geotagging by GPS or baking produc- tion and commissioning credentials onto recording media, then it will likely lead to problems. Deliver- ing specifi c metadata for processes not directly rel- evant to the immediate production can impact on turnaround time.” For Forbidden Tech- nologies chief executive


Stephen Streater: “Even outsourcing the process is a risk.”


‘The point is not to use automated processes in isolation but to augment manual


processes’ Joe Trainor, DMT


He adds: “There may be a particular


programme angle or subtle directorial objective that only logging teams close to a production will understand. If the logging team is not close to the vibe of a production – or they have a very detailed brief – then logging won’t be as effi cient as it could be.”


Automating the process While technical metadata such as camera ID and clip number are auto- matically generated already, the answer may be to employ automated metadata capture systems to extract information both from audio via speech-to-text transcription engines, and by using algorithms capable of identifying features and faces within footage. However, for contextual logging,


Clockwise from top left: Walking Dead Kill Count app is powered by Civolution’s audio watermarking technology; metadata can drive second-screen apps; Apple’s robotic assistant, Siri; (left) Friends


artifi cial intelligence can’t yet beat human input. Humans can add subjec- tive descriptors that indicate whether a shot is worth using, something com- puters can’t be trusted to do. Even Apple, with an R&D budget


the size of a small nation, has yet to properly crack voice recognition with its robotic assistant Siri. “We’ve given up on voice recogni-


tion until we see it working properly in the consumer domain,” says Kirk- aldie, who examined and discarded military-grade systems for ITV’s


www.broadcastnow.co.uk


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12