BBC R&D

Posted by Ben Clark on , last updated

Tellybox

Libby is planning our trip to show our "half resolution" prototypes in the Blue Room in Media City. Tim's been looking at previous work by our friends at UCL on VoxBox, trying to help us evaluate our tellybox prototypes with groups of people in a way that’s fun and appealing to participants, but also scientifically rigorous.

Talking with Machines

Henry, Tom, Andrew and Joanne have been working on our design research project with the Children's team in the BBC. They kicked off with a workshop where they used our Scenario Mapping technique to identify some promising areas for Voice-UI apps, and then worked up some ideas from those scenarios into storyboards with some guidance from the Children's team.

Tom has been re-engineering Teddy Ruxpin in order to put him to work for Talking with Machines and Tellybox. Ant has been looking at Google Home; both api.ai, its high-level CUI design tool and the lower-level Actions API.

Newsbeat Explains

Tim, Joanne and Barbara have been continuing to analyse data from the Newsbeat Explains trial. Barbara’s been interviewing journalists who worked on the project, Joanne’s been coding interview and questionnaire responses and writing up our findings, and Tim’s done some further work evaluating usage logs from the trial, looking for interesting patterns of behaviour that might help us corroborate our findings from the interviews and questionnaires.

Atomised Media

Chris and Lei from News Labs have been working on a tool to allow us to search news articles by Linked Data Platform tags. The work was started by R&D trainee Alan, and we've been working to add more detail to the search results and to allow the tool to work with both news and sports articles.

Public Service Internet

Tim, Olivier and Sacha met with David Moats from Linköping University’s Department Of Technology and Social Change to discuss ‘fake news’ and fact-checking, and explore ways that our work might align with David’s research interests - in particular ways of visualising and authenticating the provenance of online information.

The Secret Science of Pop

You might have seen a trailer on BBC4 recently for a documentary called The Secret Science of Pop which we worked on last year and are excited to finally be able to talk about publicly! Tim’s spent some of this week writing up our work on it for a blog post which we’ll publish soon.

Editorial Algorithms

David, Matt, Frankie, Olivier and Kate have been working on our front-end tools. The Stream Builder now allows end users to create their own searches and is being alpha tested with BBC colleagues in IRFS and at demo days. David completed the designs for the Compact View in Search & Stream - the tool that allows users to see and filter the results of their searches. The first iteration of changes incorporates feedback from our colleagues in BBC Monitoring. Fionntán has started working on “Smart Streams” considering how we might kick-start end user driven training and improvements to Streams.

On the Platform side, Josh has been making amazing progress moving our components to the BBC’s COSMOS platform - a process that will ultimately allow us to handover support of the Editorial Algorithms backend/platform to the BBC’s 24/7 Support Team.In the research space Fionntán has started on an API and front-end demo to show off what is possible with the clustering and news-space ideas – grouping news stories both by event and by topic.

Olivier and David took part in a day-long event organised by Connected Studios and the Academy, to share the tools and demonstrators we’ve built on top of our “editorial algorithms". There was a lot of good feedback which should mean more people using our tools.

Vivo-NewsLabs-IRFS Hackday

Katie and Barbara worked with members of Connected Studio, NewsLabs and CPS-Vivo (a News production tool) to arrange a joint HackDay. So far, we've had a pre-hack meeting to introduce the teams to each other and shared knowledge of the platforms and tools that they work on and with. As part of this Olivier presented the Editorial Algorithm APIs, raw tech and tools. Tim’s been raiding the World Service Archive project to generate datasets for the hackday - there’s crowdsourced and editorially checked tags for about 20000 radio programmes and we’re looking forward to seeing what people at the hackday might do with them!

Starfruit

Chris Newell has been exploring feature selection in our Starfruit tagging system, with the aim of improving performance and reducing model size.

Speech To Text

Matt and Chrissy have been building an acoustic model for end-to-end Speech-to-Text using Tensorflow and have reproduced a model from a paper and are now testing various approaches to decoding the model.

COMMA

Matt, Chrissy and Ben have been testing and fixing bugs in Comma and writing documentation. Ben has also been making Web VTT from the output from various algorithms being developed as part of the CAT project so that we can display the data alongside audio or video in the browser. He has also been adding centralized logging and metrics collection to Comma using Kibana and Grafana.

Scene Detection

Craig has been implementing the algorithm in Makarand Tapaswi's "Story Graphs" paper to detect scene boundaries and evaluate it on BBC programmes.

Other Stuff

Working with Richard Sewell, Libby has made a new version of libbybot, her webRTC-based robot avatar.

Chris was at the EBU last Monday and Tuesday for the annual Radio Hack. He worked on adding RadioDNS capability to an Android app using the OMRI API.

Tim gave a talk to the R&D engineers meeting on Haskell, Functional programming, and writing programs for humans first and machines second.

Links

  • On-Air Audio Processing: Getting the most from your broadcast audio processor - handy intro to the magic that happens in between the studio and your radio in order to make everything sound good!
  • Selfwork: How fitness technologies turn the body into an investment property - great read on the political implications of quantified-self technologies, time management and the culture of ‘big data’.
  • The music theory principle that unifies 2016’s radio hits - more ‘Science of Pop’ from Vox magazine - a great nerdy music-theory read on a recent trend in pop songwriting.