Find out who is doing the speaker selections (even if they are just shortlisting) for the conference. Try and find a way to 'meet' them (this may be online). Find if they have a blog, and comment thoughtfully on their blog. If it's a yearly conference, go to the conference and seek them out and introduce yourself. Go to a session on the topic you want to talk about, and get up, introduce yourself (it feels tacky and awkward, but comes across as very professional) and then ask an intelligent question that will help the presenter cover an aspect they maybe didn't get time to cover. Ideally not a 'gotcha' question.
The aim is that when your submission is on that persons desk, they immediately have a picture in their head of who you are, what you know, and that you can write coherently on the topic at hand.
I followed this prescription and went from nobody wanting me to doing several invited user groups per year and speaking at at least one big conference per year, and also being invited to deliver webinars with quite large audiences.
Finally, don't mess up your big chance when you get it, and beg the attendees to put in evaluation forms if they are there. Hand out candy if you have to.
"Your submission will be judged by your ability to communicate succinctly in writing, how good we think the talk sounds, what other people say about you behind your back and whether, frankly, we feel like people are going to be happy they paid money to see you yap for an hour."
It sounds a little harsh, but I think it's true. Getting to know people involved in the community or conference organization will pay off.
It's a big leap to go from a short user group presentation to hundreds of people listening to you for an hour. So you should expect to spend time getting to know people and money on attending other events in order to get there.
Of course, if you have a truly awesome new idea that nobody else has thought of, you can skip that step, but very few people do.
i am, however, off the stage and off the road by choice. couldn't be happier.