Posted by Anthony Onumonu on
Welcome to the regular update from the Internet Research & Future Services team in BBC R&D. This week: Women Driven Development and World Mental Health Day.
Women Driven Development
Misa, together with Phoebe from the CPS team in London, organised their first Women Driven Development (WDD) hackathon at Google on October 17th. They founded WDD earlier this year to tackle gender imbalance in the tech industry by providing opportunities for gender minorities and tech leaders to work together, learn from each other and create organic connections.
Photo Credit: Eddie Jaoude
From IRFS, Alicia, Joanna and Kristine also joined the hack. Over 60 attendees were split into 10 teams and worked on projects to address issues such as mental health in youth, imposter syndrome, accessibility and plastic waste. You can find more about each project in the WDD project repository.
Misa wrote a blog post to share her journey in the tech industry, how WDD came about and how they made this hackathon happen. The day was documented by Eddie Jaoude, a senior open source developer, who is passionate about raising awareness of diversity and inclusion.
World Mental Health Day
To contribute towards World Mental Health Day this year Todd was inspired to see what R&D could do to investigate the topic of how technology affects mental health and wellbeing. The idea was to use a one-day workshop/hack day to understand how the BBC focuses on mental health and if there was anything we could do to try and build better experiences. Whether that was looking at existing services and changing them to better account for mental health or prototype some new ideas to address this need.
The day started with two eye-opening presentations from Nelson Kumah (BBC Sport) and Lianne Kerlin (R&D FXT). Following this, there was a period of ideation before diving into making some prototypes. Two of the prototypes designed are highlighted below.
This was developed to address the user needs of being adversely affected by the news and that people may want to be warned about upsetting stories or avoid specific issues. To address this Chris, Alicia, and David designed a collection of ideas which allow users to filter the news they consume. Whether this is using a users mood to show only stories with the appropriate sentiment, or visually displaying the ‘mood’ of the new article giving users a more informed choice about what to read. This would hopefully reduce anxiety and distress which can be caused by the news.
Jimmy, Natasha, and Kristian prototyped an object-based media Alexa skill to guide you through a personalised meditation session. The idea was to take users preferences for BBC presenters, BBC shows, and the time available to personalise a meditation experience using content from the archives. This would allow people to take time when they can to relax with something they enjoy and focus on their mental wellbeing.
Chris was at W3C TPAC, where he chaired a meeting of the Media & Entertainment Interest Group, and attended meetings on second screen communication protocols, HDR colour, media timed events, Web Assembly, WebGPU, VR/AR, and credibility of information and fake news.
Talking with Machines
A productive sprint where we all worked together on The Unfortunates to get it into Amazon for certification. Anthony implemented APL, Joanna and Andrew edited and reworked all their designs to fit the new templates, Henry worked on the main skill (help functionality and extra questions thrown in at the last minute) and Nicky did the editorial. Nicky has also begun working on comms for the project doing an interview for Radio 4.
Nicky and Lei are working on a better more rounded version of their Guess Whom app that they built at the multimodal/voice hack. Lei has put it on Pi with help of Libby and Nicky has expanded the questions.
Speech to Text
The Data team has built a new Kaldi model with their extended lexicon. This has not affected the overall system accuracy on our test sets but we are now seeing new words appear in our transcriptions - which is what we hoped for at this stage.
The team will now increase the system's language model to reflect the context of the new lexicon better, this should improve system performance on our new lexicon.
Denise is continuing learning the Tensorflow aspect of the Speaker ID system and evaluating the embeddings from the original VoxCeleb paper.
This sprint, Tim, Chris, David and Caroline met with the Political Research Unit at Millbank to discuss our Quote Attribution database/tool. They spent a day shadowing the unit and learning about their process.
David, Jakub and Tim worked on a new web prototype focusing on exploring the wealth of BBC Sounds clips.
"It was noticeable that most programmes are segmented into shorter pieces of content (there are hundreds of thousands of them) but these ‘clips’ are quite hidden. This prototype aims to make better use of the radio clips and to make it easy for a listener to get a sequence of clips for a duration of their choice." – David
This prototype is part of a wider set of explorations we call "Enriching Discovery Mechanisms" – we're investigating means of discovering BBC content beyond personalised recommendations, focusing on sharing, contextual discovery and generative content.
This post is part of the Internet Research and Future Services section