Oversampling conversation

Says an error has occurred.  Looks like there's an extra period at the end of the link.

It's an interesting video but I think it comes at oversampling from the wrong direction.  We use oversampling because it eases the requirements for antialiasing filters, not the other way around.

You'd upsample the original waveform (which will likely already be band-limited), which is not hard, but it is time and resource intensive as they mention in the video.  However, the oversampling moves the Nyquist frequency up considerably, which allows the use of lower-order filtering.  This reduces the need for filters with fast cutoffs but also have high over/undershoot. 

I’m dubious of anything that dingdong has to say. Nothing I’ve seen from has ever suggested he has any real insight.

Who’s dingdong

Ok. I was listening more to the guy he interviewed

This is interesting. I definitely learned some stuff, especially about sample theory. I can’t think of anything I’ll change about my processing, but good to know how much is involved in designing good OS.

Thanks for posting, as I would not have clicked on one of that guy’s videos otherwise.

UA internally upsamples all of their plugins to 192k for this reason.

Oversampling merely extends the bandwidth to accommodate the harmonics and phase shift of functions that exceed the nyquist frequency of a sample rate. It's a technique for distortion prevention that's best kept out of the hands of the user no matter what the doofus on the left and the programmer on the right say in the video. Anyone can try to make their computer perform functions with infinite harmonics that oversampling will barely help to mitigate. How is the user supposed to know which operations are non-linear and they need X amount of bandwidth to accomodate the harmonics? What if the programmer writes a hard clipper into the code with higher order harmonics extending far beyond 96 khz, the limit for 192 khz? OVC-128 and Standard Clip have ridiculous 128 and 256x upsampling respectively to attempt to accommodate the infinite harmonics from hard clipping.

There are other distortion reduction techniques such as writing functions that will have not have excessive harmonics (polynomial functions instead of soft clippers, you can write a cleaner rectifier than you can build, using antiderivatives (difficult for compressors and hysteresis functions but apulsoft and newfangled audio implement them for waveshapers)), lookaheads to smooth the attack to lower distortion, frequency dependent behavior (think an API 525 but far more complex), "auto" time constants in compressors that take into account time to calculate the speed of gain reduction and restoration, and straight up brutally limiting the bandwidth of the sidechain to control aliasing like Oxford Dynamics and Renaissance Compressor did.

There are several developers, e.g. Tokyo Dawn, Goodhertz, and U-he, who oversample only as much as each function or set of them needs and then have only something like 2-4x oversampling for the audio path at 44.1 and 48 khz to prevent imd from modulating the audio and cramping in eqs but the control paths can exceed this greatly. U-he Satin has functions at 384 khz while the Crane Song Titan and TDR Kotelnikov GE can upsample internally up to 20x. There are no downsides to this despite what Whitesnake and the developer try to claim in the video because the filters are not in the audio path but the sidechain. Without high internal sample rates, the operations those processors would attempt to perform on the audio would be dysfunctional.

The Weiss DS-1 had a 2x oversampled audio path and 4x oversampled sidechain back in the 90s. The compression algorithm still sounds pretty good if a bit colored, distorted, and sometimes thwacky versus modern plugins. It's just hard to set with the two time constants and the release delay parameter needing to be set to the amount of lookahead to not release prematurely. Very primitive and more similar to a cutting lathe limiter like than most digital limiter plugin or manual or automatic holds or slowdowns in gain recovery in certain compressors to prevent noise rushes when quiet passages happen directly after loud ones.

I'm not disagreeing but some native plugins go even higher for certain functions. Their emulated compressors aren't as good as the digital ones you can get native now but most of them are older.

The discussion in this thread perfectly illustrates why I am annoyed/frustrated by the host of that show. He’s not equipped to ask thoughtful, high level questions that informed real world practitioners would have, because he isn’t one. Y’all would have made a far more compelling interviewer.

Math or Meth?

Yes, that's one use of oversampling, but the video is talking about creating frequency headroom (increasing the working bandwidth) for subsequent non-linear processing. With the added headroom, the aliasing doesn't extend back into the audio band as quickly/easily, and is removed on the downsample back to the base rate.

The video goes overboard on the FUD (ear, uncertainty, doubt) factor. Seemingly imagining that we typically use 6-8 non-linear plugins on a track, and the plugin designers may have made decisions that won't stack well with each other, etc. (But we'll happily live with vintage analog gear chains up to our level of noise tolerance, and celebrate the sound.)