What sound interpolation method is better?

KaOSoFt · Post by **KaOSoFt** » Sat Jan 10, 2009 2:14 am

Hello.

I read a topic in which the definition of each method is given but, the thing is, 8-point interpolation is the only one missing a non-technical description.

Well, it says Cubic is better than Gaussian, so, is 8-point better than Cubic (if you read the options top-down)? I tried looking around, but didn't find any clear explanation.

Thanks in advance for your support.

Post by **Agozer** » Sat Jan 10, 2009 3:33 am

If you want something that is closest to the real thing, use Gaussian.

h4tred · Post by **h4tred** » Sat Jan 10, 2009 7:37 am

so, is 8-point better than Cubic

Basically, sinc ("8 point") interpolation is better.

Here's a explanation on linear and sinc interpolation.

Post by **grinvader** » Sat Jan 10, 2009 12:12 pm

Interpolation methods use math formulas that "fill in the missing parts" between each value. The best result is obtained when the formula fits the actual sound variations perfectly.

Cubic and gaussian are both 4-points, but they use different formulas. Gaussian uses a formula that yields less errors on edges.
I haven't looked at the code to know which sort of 8-point interpolation ZSNES uses.

Note that SNES hardware uses a gaussian interpolation on its samples after they go through pitch adjustment.
There's a second point where interpolation may occur, and that's between the emulator and your sound card (resampling to fit your card's native rate, for example).
I don't exactly remember if the option in ZSNES applies to the former, the latter, or both (in the current releases, at least).