What IRFS has been up to this week: delivering new speech to text models, testing BBC Together, and 'transparent' iPlayer recommendations.
One of the biggest challenges in research and development is to get your ideas adopted by the business. Corporate history is full of tales of innovative research teams doing groundbreaking work that’s overlooked. Xerox PARC and the development of the graphical user interface is probably the most famous example of a corporate missed opportunity. So it’s good to hear that the Data Team have been focusing on the crucial 'tech transfer' part of the research cycle this week.
The Data Team’s Speech-to-text group discovered this week that their speech-to-text research is a key part of the core functionality of the newly launched Transcriptor service. Transciptor has been developed by the Newslabs team as a way for BBC staff to get any piece of audio or video quickly transcribed. It's likely to be a big hit with time-pressured programme makers who need to quickly scan footage for quotes. It launched this week to 6000 journalists and the team are looking forward to hearing feedback from their new user base. The team have also been working with Platform Media Services to make sure the latest and most accurate version of our speech-to-text system is rolled out to the business for March, so another tech transfer tick in the box for them.
In a similar vein, this week the Recommendations Team have been looking at better ways to compare research and live production models so that improvements we make here in R&D can be transferred over to iPlayer users even quicker. The Recs team probably hold the record for the quickest tech transfer of any of out teams, with improvements to recommender models made by them often being seen by end users in a matter of weeks.
The Data Team’s NLP group is continuing to explore the use of Artificial Intelligence to identify abusive and offensive behaviour in BBC online comment forums. It has recently approached a number of academic groups and would be interested to talk to others working in this area.
Finally, this week we say our farewells to Denise Bland, who is retiring from R&D after 20 years. Denise has been with the Data Team for the last 6 years and most recently has spearheaded our work on Sentiment Analysis of audio and video. The whole team (and, of course, the wider department) would like to thank her and wish her well on her new adventures.
In the Internet and Society team, Libby has been supporting the testing of BBC Together as Kristian moves it to Cosmos, BBC's AWS platform, based on lots of work by our very own Jess. We've had some exciting promotion with Drag Race UK and on the One Show(!) and this work enables us to scale it if necessary, and supports automated deployment.
IRFS had an open show and tell within the BBC and Alicia demoed her Impostor Syndrome Challenger chatbot. She's been sending the chatbot and survey to potentially interested people and finding places to advertise it to get some good feedback. You can give your feedback here if you're interested in techniques to challenge your imposter syndrome.
David has been iterating his BBC introducing design/wireframes for our work with the Data team on a prototype machine-learning-based music matcher, working with Kristine, BenH, BenC and Alicia. He's also been continuing the interviews with BBC journalists about their views on the Full Fact fact-matching tool.
Chris has been having conversations with people across the BBC on Google's Privacy Sandbox, which aims to remove third-party cookies from the web, to understand any potential impact these changes in browser behaviour may have for the BBC.
Tristan has been writing up his, Alicia and Holly's "transparent" iPlayer recommendations testing, and finding ways to summarise our last six month's work on understanding and controlling machine learning.
Libby and Henry have been thinking about tools and process for futuring and some possible workshops. We're reading How to Future by Scott Smith with Madeline Ashby, and Future Ethics by Cennydd Bowles.
Over in the Interaction and Prototyping team, Mathieu and Andrew have been refining the multiplayer adventure interface ahead of of internal testing, focusing on attention division issues. Additionally, we’ve made some open source contributions on a plugin related to the prototype and worked on integrating sound in the adventure.
Anthony has been documenting the work done for the voice-first format project. He's been also thinking on how to improve our ways of working and tools we use.
Barbara has been mainly dividing her efforts between the Diversity & Inclusion Working Group - working on the objectives (OKRs) for 2021 - and the Connected Cars Working Group. She's also planning the evaluation of the synthetic voices study for which a good data analyst is needed. (Contact Barbara if interested)