FIAT/IFTA World Conference 2019

Fewer people, more content: it is clear that VRT’s archive department is under a lot of pressure to process all incoming content. Based on the buzz, AI could be the solution. But is it really and are there other ways to facilitate the archivists? To get more insight in the new upcoming technologies, VRT started an exploratory project divided into two tracks. The first track tried to capture already existing production metadata. The second one investigated how AI could enrich content.

1. In the creative process of making media getting all the relevant metadata to the archive is not straightforward and a good understanding of the production process is crucial. To decide on useful integrations metadata flows and, more importantly, problems in these flows were listed and prioritised. Two cases were implemented. Sport archivists, for example, could benefit by using time coded game information produced for the online sports live stream as metadata. The implemented integrations rapidly introduced time savings, but also created metadata that’s less optimized for archival purposes. Furthermore, we will need to be alert to keep up with the ever-changing processes.

2. AI gets its information from the content itself and doesn’t necessarily require input from production flows. Because the AI possibilities are numerous, we started off with a comprehensive market research and some minor but inspirational tests. This lead to two use cases, speech recognition and text analysis more specifically, which we investigated in more detail. Results were noisy and within the small period of the POC no time gain was received. During the experiment we realised AI will need a lot of effort in the beginning, although it did already show potential in, for example, subordinated cases.

 In the end, each track had it advantages and disadvantages and our future archive will need to be flexible and create expertise in different methods in order to be able to choose a good strategy for each use case.