BBC R&D

Posted by Chris Newell on

This sprint we've explored mood filtering, synthetic voices and Audio AR. We also held a joint hackday with the BBC's DataLab team and joined a W3C workshop on Web Games.

News Mood Filtering

The Better Internet team has completed their sprint on the News Mood Filter project which allows users to avoid specific topics or negative news by adding warnings and masking selected news stories. The project emerged from a hackday exploring mental health issues which identified that some people avoid news because it's depressing and makes them feel helpless. The team think that giving people more control could mean that they're more likely to read the news.

The prototype shown above, uses a Chrome extension and has allowed the team to demonstrate and discuss the concept with various groups inside and outside the BBC. The updated prototype allows users to specify how long a topic should be muted and includes a new "mood wheel" which enables users to indicate their current mood. The information is used to personalise their service whilst not hiding the existence of any content or making it completely inaccessible.

Synthetic Voices

The Synth Voices project is exploring what the BBC should sound like, in services which use synthetic voices. The project is a collaboration with the BBC's Voice + AI team, the BBC Science Unit and Salford University.

The project has been acquiring sample voices and exploring the extent to which they can modify and control them. They've also started to consider the voice properties they want to explore, such as:

  • gendered and gender neutral voices
  • celebrity voices
  • accented voices and regional variations
  • age variations
  • emotional voices

 The goal is to be able to reflect the diversity of the UK as much as possible.

Audio AR - Smart Headphones

For our work on Bose Frames (one of the first commercially available Audio AR products) Anna and Valentin from QMUL have been demonstrating a series of multiplayer games. They've also written a blog post and paper on Collaborative Experiences in Audio AR for the EuroVR conference.

Meanwhile Nicky and Henry have been in co-creation mode with sound designers Ben and Max Ringham. They've scripted and recorded a four scene drama which plays with binaural sound, voice interaction and gesture. 

Joint IRFS / DataLab Hackday

Last week IRFS held a hackday with the BBC's DataLab team, splitting into groups to explore four topics:

  • Disinformation: can we counter disinformation by providing different perspectives on the same topic?
  • Evergreen: can we automatically identify content which is timeless, rather than time-limited?
  • Pipelines: can pipelines tools, such as Apache Airflow and Kubeflow, help us train and deploy machine learning models?
  • Trends: are the trends on BBC services similar or different to the trends on external services, such as YouTube?

Although the hackday was too short to make substantial progress on these topics, it achieved our aim of building relationships and exchanging ideas.

Content Analysis Toolkit

The Data Team have been adding the finishing touches to the demonstrator for their Content Analysis Toolkit. The demonstrator will illustrate how it can be used to create searchable transcripts of radio and TV programmes and provide a showcase for the speech-to-text, face recognition with speaker ID technologies.

Standards

Chris attended the recent W3C Workshop on Web Games. This explored new technologies and APIs that could make the web a more powerful platform for games, providing exciting opportunities for BBC services.

Topics