Hi! I switched at around 14 seconds too.
It won't take long to get from your current Roux speed to your current CFOP speed (definitely not as long as it took you with CFOP), but after that it's not really much faster, unless you know where you are slowest. If you know one step is worse than the others, then you can focus on that and improve it, and that will help you get faster.
Firstly, be efficient. Know how to solve all the EO cases optimally (which isn't as hard as it sounds), and learn how to influence ULUR during EO. You can do this by changing some U moves into U's and M's into Ms. Since the ULUR colours are easy to see during LSE, you can track them relatively easily. I find doing lots of slow, untimed solves was helpful for this.
You want your F2B to be roughly half your solve, so 4.5 is a good time to aim for. Maybe 1.5 seconds for FB and 3 for second could be good.
As PDF said in the post two above this,
algdb.net is the best website for algs. Most algs from other websites are on there so you don't really need to look at any other sites.
Efficiency is important during your first block. At first, try to plan just the first square during inspection. As you improve, try to influence the pair - again, untimed solves are good for this. You eventually want to be planning the full first block in inspection, but don't worry about that for now.
For second block, again, you definitely want to do slow, untimed solves. I found it helpful to do slow solves where I forced myself to be turning constantly - not completely focussing on effiency, but still looking for cancellations and other tricks I might not see in a speedsolve.
Good luck!