A computer can pick out speech even amid cacophony

November 26, 2008
Schematic diagram of SHoUT

(PhysOrg.com) -- Using a recent development in speech recognition, it is possible to search through television news programmes provided the recognition system has been trained beforehand. PhD candidate Marijn Huijbregts from the University of Twente (Netherlands) has, however, taken things even further: he has developed Spoken Document Retrieval for audio and video files that the speech recognition system has not yet been trained to deal with.

This version of speech recognition works well even if there is a great deal of unexpected background noise. Huijbregts received his doctorate from the Faculty of Electrical Engineering, Mathematics and Computer Science on 21 November.

Information can be retrieved from text very quickly using, for example, an index in a book or a search machine such as Google. However, it is much more difficult to search in audio and video files, as they do not have an easily searchable index. You can use speech recognition to simplify this process as most of the information in audio and video files comes from speech. By recording via speech recognition, you can transform speech into text. To do this, you need a Spoken Document Retrieval (SDR) system; this makes it possible to search directly in audio and video materials, just as if you were searching in ordinary text documents. In other words, a sort of Google for audio and video.

Evening news on television

The Human Media Interaction group at the University of Twente had previously developed an SDR system for an evening television news programme. Search terms could be used to look for specific topics, the system being specially trained using newspaper texts and 20 hours of news programmes. The SDR for the evening news programme worked well because, in that situation, it was more or less known what was going to be said and there was little background noise. If you tried applying this system, without any training, to other video files, it did not perform well. Huijbregts then wondered whether he could develop a SDR system for which almost no training data would be needed, but which could nevertheless deal with unknown audio and video files satisfactorily.

SHoUT

With unknown audio and video files, it is not clear beforehand what is going to happen: who is speaking, what is being said and what sort of background noises are present. Huijbregts therefore developed an SDR system that was robust enough to deal with these unknown situations. It is called SHoUT (this acronym corresponds to the Dutch version of ‘Speech Recognition Research at University of Twente’). SDR can be described as robust if it can deal with all audio and video files under all sorts of circumstances, such as background noise or if people are not speaking clearly.

SHoUT is divided up into three stages. Firstly, the system distinguishes between speech and other sounds. For example, background music is filtered out from speech. Secondly, the system identifies different speakers and gives them labels. Then finally the automatic speech recognition takes place: the system transforms speech into text. You can now search the text file for relevant topics using key words, just as Google searches through text files on Internet.

The first version of SHoUT is already available, but Huijbregts is developing it even further. SHoUT and other demonstrations of SDR systems can be found on the website of Huijbregts (wwwhome.cs.utwente.nl/~huijbreg/).

Provided by University of Twente, Netherlands

Explore further: Team created techniques to analyze thousands of hours of NASA tape

Related Stories

Recommended for you

FCC votes along party lines to end 'net neutrality' (Update)

December 14, 2017

The Federal Communications Commission repealed the Obama-era "net neutrality" rules Thursday, giving internet service providers like Verizon, Comcast and AT&T a free hand to slow or block websites and apps as they see fit ...

US faces moment of truth on 'net neutrality'

December 14, 2017

The acrimonious battle over "net neutrality" in America comes to a head Thursday with a US agency set to vote to roll back rules enacted two years earlier aimed at preventing a "two-speed" internet.

The wet road to fast and stable batteries

December 14, 2017

An international team of scientists—including several researchers from the U.S. Department of Energy's (DOE) Argonne National Laboratory—has discovered an anode battery material with superfast charging and stable operation ...

1 comment

Adjust slider to filter visible comments by rank

Display comments: newest first

NanoStuff
not rated yet Nov 30, 2008
Youtube becomes self aware.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.