First, if you want to generate a voicefile for other players then those old Archos ones, you need to encode you wavfiles with rbspeexenc (you can find it in svn under tools) and not as mp3.
I dont know if the exe from the wiki is still uptodate, so better build your own from current svn.
The string - id list is normally generated out of english.lang, dont know the exact details, but you could look how its done in voice.pl or in rbutil. You also need the list of all features (look at rockbox-info.txt on your target).
Perhaps it would be easier to use the buildsystem to generate the voicefile, than todo it all by hand. You could try to configure the buildsystem to use a POOL for the voiceclips, run it once, replace those clips in the POOL directory, and run it again.