What Happens During Downsampling?

[seanmccoy]

Hey, all. I know there are other threads discussing the merits of high sample rates, but until now I've stuck with 48 and 44.1. I have two upcoming projects that will originate elsewhere at 96 kHz, so my hand is being forced into that rarefied air.

I have a pretty good handle on the technical side of bit reduction and dithering, but really don't know what happens during downsampling in Pro Tools, and how the higher quality is retained during the process. Does the audio stream internally and get re-recorded from the original, or is it simply a mathematical algorithm that figures it out?

And is there a consensus method, chain of events, or recommended plugin(s) that will ensure the highest quality resultant files? Any basic explanation, or external reference to one would be appreciated.

[JFreak]

First of all, you would not dither unless you go to lower bit rate. Second, when you SRC, there is no such magic as "converting from 96k to 48k is cleaner than 96k to 44.1k because 96/48=2". When you SRC, the actual waveform is reconstructed and then that is resampled to sample rate of your choice.

You will lose all frequencies outside Nyquist Theorem, as in 96k session may have frequencies up to 48kHz (but likely will not) whereas 48k session may have frequencies up to 24kHz. Most likely somehwere along the signal path frequencies have already been filtered to 20Hz-20kHz

[Darryl Ramm]

Think you meant: First of all, you would not dither unless you go to lower number of bits/sample length.

[JFreak]

Yeap, pardon my french

[seanmccoy]

Yes, I understand the difference between bit depth reduction and SRC. So by "reconstructing" the waveform it sounds as though there is a virtual re-recording at the new sample rate, so that none of the advantages (theoretical or otherwise) of the original higher rate are lost.

Practically, when working with high rate source material, then, I would assume it would be best to stay at the higher rate until the very final stage, whether that's mixing for film/video or mastering music for CD or streaming (similar to the way bit depth is handled when 16 bit files are needed), and that the final SRC should take place as a separate process before the final bit depth reduction, yes?

[JFreak]

Yes and no.

If the material only contains frequencies lower than 24kHz, there is no harm in SRC to 48k but if there are some bat material which you cannot here but want to use, then don't do it.

[JingleDjango]

For a better overall understanding of digital audio and how to retain fidelity from generation to generation, Bob Katz' book Mastering Audio is an excellent reference.

[JFreak]

^^ this

And by audio reconstruction I (and mr.katz) mean that SRC is not done sample by sample but instead the waveform is recreated as it is, and then that waveform is resampled to desired sampling rate, obviously after filtering out impossible frequencies.

[seanmccoy]

Thanks for reminding me about Bob's book, which I have and have read, but clearly need to revisit the sections on sample rates.

[seanmccoy]

And while we're (still) on the subject, any objective opinions on how Pro Tools' SRC compares to other options, such as Izotope's?