FEED Issue 11

35

ROUND TABLE Content Management

people to check output and correct it where necessary. AI may be an improvement – but we’re not ready to just push the button and see perfectly rendered metadata created from unstructured data. We often see overwhelming amounts of data being produced by AI routines, which means you need a machine-learning step to create more concise output – to train machines to spot unknown faces in the archive, to give an example. We probably need to manage the expectations around AI, particularly with owners of rapidly decaying archive content. It’s not going to magically catalogue their content for them, but it may offer cost effectiveness and improve accuracy along the way. KEVIN SAVINA: You first want to automate the digitisation process as much as possible in getting the catalogue ready and orchestrate the human cataloguing aspect of it. That gives you a first level of creation of the archive. That can be done by workflow tools that are included in workflow and media asset management systems. On top of that, you can use AI- based technologies to enrich the archive. AI technologies can help catalogue the content – speech to text, computer vision and keyword detection. Those technologies

back-end catalogue, you can expose it through multiple revenue-making avenues.

GOODY GROUP: In what ways and to what extent can new automated and AI-driven technologies help us to organise and catalogue all this extra material? VIVEK KHEMANI: As mentioned earlier, AI can help in tagging at a very granular level of data. One can tag metadata at frame level as well as within a frame. IAN MOTTASHED: AI and automation are great, but not a silver bullet. Let’s take a well-established concept like speech to text. We probably all know by now that it works best when people speak clearly, one word at a time, with no background noise. How much of the footage will give us those conditions? If we’re dealing with thousands of hours of imperfect audio conditions – this is a gnarly old archive after all – we’ll still need

feedzine feed.zine feedmagazine.tv

Powered by