[Sesse__]

> If the rate isn't an even multiple, it's worth using a tool that can dither; dithered resampling artifacts are less abrasive than undithered...

This seems to be mixing up two things; proper interpolation and dithering.

If you have limited bit depth (in practice, 16 bits or worse), you should pretty much always dither, ideally also noise shape. This is independent of the interpolation you're using; having a rational relationship between the original and downsampled signal makes some of the implementation a bit easier, but even for something like 48000 -> 24000, you'll end up with effectively a float signal that you need to convert to your chosen bit depth somehow, and that should be done better than just truncating/rounding.

And even for interpolating between two prime rates, or even variable-rate interpolation, you can and should get great interpolation (typically by picking out polyphase filtering coefficients from a windowed sinc of some sort).