Speech compression is a technique which reduces the length of a speech recording but retains the important content. We have experimented with a variety of techniques for applying this compression, of which we have several examples:

• Plain speedup – here the clip is played back at a faster rate whilst the pitch of each speaker is kept constant.
• Variable playback rate – here the intention is to mirror how humans naturally alter their speech rate whilst speaking. The clip is played back here at a variable rate which makes it easier to understand.
• Insignificant word excision – here we use simple Information Retrieval measures of word importance and build a clip which contains only the most important words. It is understandable but far from natural speech.
• Summary excision – here we use the previous measures of word importance to derive measures for longer parts of speech (Utterances) and build a clip which contains only the most importance utterances.

User Benefits

Users may be able to experience a meeting in less than real time.

Research Challenge

This area of research is very challenging

Status of the Research

This research is ready to be evaluated.

Contact for this demonstration

For more information, contact Simon Tucker, University of Sheffield