About

GlobalVox is an end-to-end system for processing and browsing audio news data. This fully automated system brings together our recent research at the University of Edinburgh on audio scene analysis, speech recognition, summarisation, named entity detection, geolocation, and machine translation. It currently automatically processes various freely available BBC news radio podcasts.

The graphical interface on the front page allows users to visualise the distribution of news content by entity names and story location. This currently aggregrates news over the last seven days, although you can also look at data from past months. Named entities and locations are rendered according to frequency of mentions (size) and automatically determined sentiment (colour). A representative quote for individual news programs is shown on the right panel. Clicking on these quotes leads to full automatic transcripts and translations. Browsing of news events is facilitated through extractive summaries.

This demo was first conceived and prototyped at BBC NewsHACK III ('Automated International News Services of the Future')

The original GlobalVox demo came together from work from various projects, including:

It is was also the inspiration for the EU Horizon 2020 project SUMMA that some of the globalvox team are currently working on.

For more details see:

Awards

  • Best end-to-end multilanguage tools, BBC NewsHACK III
  • Show and Tell Prize (popular vote), Interspeech 2015

People