If you check out the proposals for the 2010 GSoC in the wiki, you'll see that i added "pocket jeeves" as an entry a couple days ago. I built pockt sphinx on Maemo 5 and it works: it will recognize very accurately from a small library of useful command phrases i built into a language model. There's a big step from working compnents to polished product, and one of the major major handicaps a project like this has is that supporting all the languages the product line covers is prohibitively expensive (where time=money). If only English had to be supported, i think you'd see this in the core software...but it's not that simple. In other words, this is probably going to have to come from the community and is probably not going to support all the possible languages. It will require extensive dbus and other development skills and a lot of knowledge of maemo internals to do right (=context-based).

In summary, i would like to have handsfree command mode with voice feedback (ie pocket jeeves) but don't have the huge amount of free time doing it properly would require. I'll see if there's a brainstorm covering this and if not create one in the next few days, using the same ideas i already developed for the GSoC proposal and for my own project wishlist.