Imagine you’re a TV news journalist. You and your camera crew have done live hits all day and file a comprehensive video report, wrapping up all the elements of the story for a prime-time news bulletin. Then you wait for it to be published on the Web site, where it will probably find a different, oftentimes bigger audience.

But you’re left waiting for it to be processed by someone else who manages the videos on the Web site — someone not as close to the story who needs to clip the videos out manually and key in metadata to make them discoverable.

At Mediacorp, we’ve started using Artificial Intelligence to speed up this process, with an eye to getting more news video reports and interviews online in a more timely fashion.

A smarter approach

We introduced a project called AI SmartCut, which allows us to train an algorithm to recognise packaged video reports and interviews with analysts or studio guests in a bulletin. These videos are clipped out automatically from a 24/7 livestream of our channel, CNA.

We use voice recognition, scene detection, and natural language processing (NLP) to identify news video stories and interviews and feed them into our content management system (CMS) for easy publication within minutes of airing.

We have fed the AI hundreds of hours of speech by our news anchors and journalists, and it can recognise audio patterns such as programme music.

It also has automatic transcription and keyword tagging, so the videos come with full metadata when pumped into our CMS.

The initiative, now in the final stages of testing, has shown promising results. We started out with 30%-40% accuracy nearly a year ago, but now we’re close to 80% accuracy with aggressive data validation from our journalists and refinement by AI engineers.

When AI SmartCut is rolled out fully, we expect to be able to dramatically increase the number of news video reports online, not just for CNA’s Web site but for our YouTube channel as well. People who visit our Web site will be able to see all the insightful field reports and expert interviews they might have missed.

AI SmartCut will give us more video inventory for online ad impressions and offer readers a well-rounded news experience, as not everyone wants to read lengthy articles back-to-back.

A bright future

By training AI to handle the manpower-intensive, mundane task of clipping out news video reports, we aim to free up our journalists to do more work that adds value.

We hope to do away with the tedious work of scrubbing through a bulletin recording, matching video reports against a TV rundown, clipping out a video for export and then typing in key search terms for search engine optimisation for each video.

More importantly, we think this initiative could have wider lessons for the industry, as it will help make TV news less ephemeral.

Appointment viewing has been dying for a long time. It’s time we make TV news coverage future-proof by making it more modular and primed for an online audience. The possibilities are endless for broadcast coverage to be narrowcast and more personalised online.

Ultimately, we aim to give all the good work in TV news a longer shelf life and the attention it deserves across all platforms.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .