Ah yeah, I had a feeling that was it. The problem is that Microsofts own DirectSound audio renderer for DirectShow doesn't handle fractional frame rates properly, so while the sound works fine at frame rates like 25 and 50 (PAL), the buffer doesn't get filled properly when using a frame rate such as 29.97 or 59.95 (NTSC), resulting in the audio crackling and popping. Do note however that only "Output audio to desktop" is affected by this, directly outputting the sound to the stream has always worked properly.
The noises sometimes go away after a little while, but the sound still desynchs after a while (usually a few hours), and you have to refresh the capture source. We're hoping to fix this in the OBS rewrite with our own custom audio output that isn't plagued by stupidity like this. The WaveOut renderer is only a workaround for this issue, because it does have some drawbacks. If you adjust the volume for the capture device, this will also affect the volume of the desktop audio being captured by OBS, so do keep that in mind. Mic/aux is unaffected by this.