did some research on ocr and tts-engines. while google provides us with a perfect open-source-ocr (did set it up in easx-debian) i couldn't find a state-of-the-art speech-recognition-system.

the following open-source-project was my starting-point:
http://www.simon-listens.org/index.php?id=122&L=1

it uses a japanese continuous-stt-engine, which seems to be equal/better than most commercial ones.

however it is difficult to get all the necessary sounds/statistical data. though there are open-source-projects which are collecting these.

please keep us up-to-date if you have something working on desktop or N900. i use windows vista's speech-recognition within a virtual machine in ubuntu (just out of interest for stt).

i never really used any stt-system on my desktop/mobile for productive purposes. even on my phone long-pressing a key was far superiour. but thats just my personal experience..