RadioReader is an experiment that lets journalists find, extract and publish audio clips from incoming radio inputs.


How can we help journalists find and extract audio clips from radio programmes more quickly?

Radio programmes often contain sound bites that could be repurposed for online news and social media. The difficulty for journalists is locating and extracting relevant clips from the archive — made even more difficult by the fact that the archive isn't updated with live material until about an hour after a broadcast finishes.

RadioReader is a prototype that attempts to solve the problem by feeding a radio stream directly to a text transcoding service such as Kaldi or IBM Watson. These services then return a transcript that is directly linked to the associated audio content. Journalists can then search for words or phrases and retrieve audio clips by highlighting portions of the transcript.

If we then feed the media and the transcript into a tool like OCTO, we could allow journalists to quickly extract and publish an audio sound bite to social media or alongside a news story on the BBC website.


  • We established there is demand for this kind of tool.
  • We’ve taken a lot of the ideas behind Radio Reader and worked on them in other projects.


Love data and code?

We'd like to hear from you.