View Full Version : About deriving 48 Khz impulses from 96 Khz ones

01-21-2005, 03:34 PM
(copy from my post at NoiseVault)

Next week I am going to sample a good hardware box. My intention is to sample it at least at 44.1 Khz and 48 Khz rates (24 bit).

However, I would also like to get back to this machine to re-sample it entirely at 96 Khz and to create a good collection of surround reverbs as well.

I am now first starting with true stereo and mono-to-stereo recordings (thus 3 stereo impulse files per program).

My question now is: will there be ANY audible difference between recorded 48 Khz impulses and downsampled ones (from 96 Khz recordings?). If not, I can better record directly at 44.1 and 96 Khz (saving me at least 1/3 of my recording session time).

Has anyone ever experimented with this?

I don't want to try this with the uneven division from 96 to 44.1, neither do I want to record at 88.2 Khz (because IMO nobody seriously works with that sample rate).

As for a set of "light" 16 bit versions: I guess I should just dither down my 32 bit working material to 16 bit. I see no point in setting all equipment to 16 bit and then re-record everything again, just to create a set of "light" versions of the "real" 24 bit impulses. Or am I wrong here?

Thanks for any (early) reflections ;)

Doug Marshall
03-02-2005, 02:59 PM
I have read in other forums that there is no quality difference between an original 48K recording and a 96K recording resampled to 48K - because the sample rates are multiples. No interpolation is required converting from 96 to 48, therefore, no inacccuracies are introduced that wouldn't be native to the 48K sample rate anyway.

03-07-2005, 09:49 AM
In theory there's no possible interpolation error downsampling by a factor of 2. And the opportunity to gain precision with look-ahead smoothing in software. And maybe some marginal improvement since higher res converters usually have better clocks.

But the largest practical differences have to do with the anti-aliasing filters. With 48K sample rates the Nyquist limit is 24K. But a filter at 24K in a 48K converter would require an infinitely steep cutoff which doesn't exist in the real world. And in practice, the steeper a filter gets, the more expensive it gets. So a typical bargain basement 48K converter might use an 18 or 24 db/octive filter with a -3 db point at 16K. Where a primo converter might put a 42 to 48 db/oct filter up above 20K.

As a bonus, higher frequency low-pass filters are cheaper too. So 96K converters are a win-win deal there. But the issue does shift to the quality and performance of your downsampling software. Which doesn't seem to be a big deal with all the high quality software to choose from, some of which can batch process lengthy conversions if necessary.