There are a number of open source speech recognition engines available some of which it may be possible to port to maemo. Probably the most suitable would be PocketSphinx (http://cmusphinx.sourceforge.net/html/cmusphinx.php). The problem is not so much the engine as the language translation files.

Other relevant projects are:
http://www.voxforge.org/
http://www.ece.msstate.edu/research/...ech/index.html
http://julius.sourceforge.jp/en_inde...=en/index.html

A lighter option is voice command (http://perlbox.org/) instead of recognition.

I had investigated this with a view to building something for maemo but currently don't have the time. Perhaps it will be helpful to someone else who fancies a go.