BBC R&D

Posted by Tim Cowlishaw on , last updated

It's a great week to have picked for the disco team to write weeknotes, as we can finally talk about the project we're doing with BBC4 in a bit more detail. We've mentioned it in the weeknotes before, but what we haven't been able to mention is that our work on analysing the BBC archive is going to be used in the BBC Four programme Made by Machine: When AI met the Archive, to be broadcast on 9pm on Wednesday 5th September and available on iPlayer!

This has been a massive project, delivered in a very short time, and our first time making content for broadcast, which has been a great learning experience. As well as being a fantastic opportunity to explore the creative application of Machine Learning for programme making, it’s also been a great test of using some of our other work for broadcast, especially all the Data Team’s work on COMMA and CAT.

We’ve got a more detailed blog post planned on the project, in which we’ll be able to share all the details of how we made the programme. In the meantime, massive congrats to the project team: Jakub, David, Caroline, Amaya and George, as well as Daniel Chavez Herras, our PhD Intern from Kings College Digital Humanties dept, whose input was indispensible.

Elsewhere on the Disco team Kristine has been implementing her Radio Station Themed BBC Introducing playlist audio player in React. We also welcomed back Matt Spendlove for a short while - he has built a survey application that will be used to evaluate participants viewing / reading habits within the BBC platform against their perceived behaviour, in our collaborative project with the University of Manchester. Finally, Chris Newell has made improvements to the spotter index used in the new “mentions” tagging function of our Starfruit tagging system and has created some test data to allow a quantitative evaluation.

Over in the data team, Matt has been tracking down the sources of error in our speech to text pipeline. Misa has been learning about tensorflow and has been implementing a deep-neural-network-based Voice Activation Detection component. Ben has been getting Nick’s Speaker ID code up and running, and evaluating how it performs for different tasks such as speaker ID, clustering and segmentation.

The experiences team have had loads going on - The NewNews team have been preparing two of their prototypes for testing at scale on the live news site, with help from our friends over in News Labs. Libby was on the radio talking about the Better Radio Experiences project, and is preparing for a Taster trial of our Tellybox prototypes. The ‘Talking with Machines’ team welcomed Nosa Eke, from the BBC Voice team, who will be working with us on The Next Episode on narrative structure and writing.

Finally, our amazing Joanne left our team last Thursday - she’s off to lead the new Service Design team at Hackney Council. She inspired each of us to experiment with different techniques in ideas generation and prototyping but most of all, she really kept us focussed in doing things that matter to our audience. To show how much we appreciate all the work she has done in R&D and the legacy she leaves, we have created a Joanne Moore Award for Service to Users that we’ll give every year, and Joanna designed these amazing stickers that capture key Joanne moments. There were tears, laughter and a final game of Avalon!

Joanne Moore - lunch sticker

Joanne Moore - craft sticker

Joanne Moore - analysis sticker

This post is part of the Internet Research and Future Services section