I spent the last two days fighting with that, and it's two minutes after posting that I find the solution...
It was in the previous thread
For all the people like me lurking forums to find answers: CODECS!
I forced my spa3000 and voxalot to use the ulaw codec, and only that one, and it now works. I guess that when the codec negociation occurs, it actually matters who initiate the call.
Now, I'll try to reduce that echo and see if the quality is good, or if I should try a different codec.