You should consider the consequences of only working with 3.3V.
For one, with 3.3V / 200K is only 16.5uA which is a pretty tiny current. At the very least it could take a long time for that to charge up the 10uF filter cap. You can afford to burn a few hundred uA. So using 10K / 10K divider gives you 165uA.
Two, because 3.3V is fairly limited headroom, you might consider attenuating the input if the source is a high level signal. So for example a 10K series resistor with a 3k3 to V/2 would give you -12dB attenuation.
The 5K1 input impedance is a little low. A typical line input is 10K. So adjusting the resistors in the aforementioned attenuator would increase that a little. And if you do increase it and the intent of the 680nF was to roll off the LF to gain headroom (which is not unreasonable) then you can decrease that coupling cap accordingly. But if you want full-bandwidth you should just attenuate more and let all of the LF in by increasing that coupling cap.
You might think that attenuating more would limit noise performance but realize that if the USB codec is vaguely good, it's noise floor will still be well below whatever the source might be.
Also, note that you are inverting the signal. You could re-arrange your filter to non-inverting (or rearrange the input to inverting). Or not.
