Stop oversampling everything! (plus, higher sample rates have a reason to be)

[ampetrosillo]

We all know that our ears cannot hear beyond 20kHz (and even approaching 20kHz is hard for most adults). So, the first reaction to working at higher sample rates is that it is needless, and after all we can always oversample when needed. But that's not completely true.

Even though the matter may be mostly academic, to completely shun higher sampling rates is misguided.

Higher sample rates have two main advantages:

reduced aliasing: this matters most at the processing stage, when using non-linearities. Having double the bandwidth means that plugins have essentially up to three times the headroom before aliasing shows up in the audible range; this is something oversampling can deal with, of course;

reduced phase distortion: the brickwall filter close to Nyquist will either be smoother (so easier to design with less noise and distortion, if converters run natively at the target sample rate, which is rarely the case today, since it's mostly ultra-high speed converters that are essentially sample binning to achieve the final codification) or they will be moved farther away from the audible region. In both cases phase will be less affected in the passband, since filters affect the phase well before the stopband. Phase is essentially inaudible in most cases but on some specific signals it can be; also, some people claim to be able to hear phase also in the treble region (phase is only commonly audible in the lower bass region). In any case, this is right at the input stage, when digitising the analog signal (or when converting, say, a 2MHz low-bit conversion into the final sample rate @ 24bit - the only difference being that filtering is analog in the first case, digital in the second).

Furthermore:

converters may sound different according to the sample rate they are run at, so recording something at 96k may sound different to recording it at 44.1k; this is independent from oversampling since it happens earlier in the chain;

resamplers may affect the signal significantly, although this is extremely rare with modern high-quality resampling; still, when you oversample, this happens twice each time;

certain plugins do not compensate adequately for sample rate and may have different tone at different sample rates; this makes it so that using the DAW-provided oversampling feature in Reaper, for instance, isn't as straightforward as you'd expect.

If plugins offer oversampling as a feature, this may seem to be a way to avoid the reduced aliasing but you will incur certain artifacts (usually pre-ringing, as oversampling is almost always linear-phase). They may be inconsequential, since our ears are not that sensitive to transient response in the treble region (where the pre-ringing will be most pronounced; pre-ringing is the linear-phase equivalent to phase rotation/distortion in the minimum-phase world), but pre-ringing may smear transients and if applied repetitively across plugins (say you have some track running through fifteen oversampled plugins, from start to end including submixes and the master bus) the effect may become audible enough - which could also become some sort of sound aesthetic, hey?, but it's better to know about it. In other words, oversampling is not the cure-all people may think it is, and comes with its caveats.

Ultimately, I believe that if you can track, mix and master at one rate (let's say 96k, which is low enough to ensure good converter performance) and limit oversampling to pathological cases (heavy distortions) and use minimum-phase ultrasonic filtering here and there, especially on submixes (so ensuring a mix of linear-phase and minimum-phase antialiasing and minimising phasing issues across tracks sharing some common mode, eg. drum tracks, live band recordings) you can achieve the highest quality you can; it is all very subtle though and possibly only systematically detectable by the most particular of audiophiles, and I suspect it may at most feel slightly different to the usual strategy of working at 44.1/48k and oversampling everything or close to it.

Exporting at 96k then resampling (with a high-quality resampler) at 44.1k should not affect the sound much, though. You're just filtering once and this should have negligible effects. The native OS resampling method may not be up to par, of course, so when playing back files without switching sampling rates may sound different according to the sampling rate.

So in the year 2022, with heavily multi-core CPUs, relatively cheap (and plentiful) memory and storage, and sometimes GPU offloading, unless you're using older hardware (like me) you can and should just use higher sample rates.

[Trancit]

No!

[Modular Manfred]

No

[plexuss]

Yes.

I've been doing this and advocating it for years. If you can work at 96k and it sounds better at 96k then use 96k. Some plugins dont work properly at anything but 44.1k because the dev didnt code properly in which case there may be no way to work at a higher res. When everything works in 96k, I mix and master in 44.1 and render to 96k and it often sounds superior to lower rates.

[Trancit]

nevermind... wasted already enough of my time...

[plexuss]

nevermind... wasted already enough of my time...

Good decision.

For those that can hear the difference of working at higher rates and feel it sounds better, this may be of interest:

[Aiynzahev]

I noticed that sometimes synths sound brighter and clearer at 96K. I did a 44k/96k mix down recently on a project, all compressors, VST instruments, EQ's everything, and definitely the 96k project was more relaxed, clearer and cleaner sounding. It's not just aliasing, I noticed some plugins just have a different sound and behavior at 96K, like some distortion plugins for example.

[Ploki]

nevermind... wasted already enough of my time...

Good decision.

For those that can hear the difference of working at higher rates and feel it sounds better, this may be of interest:

or just don't use distortion plugins that aren't oversampled.
like, any plugin made post 2012 by a decent company

[astralprojection]

Not quite. You should use a high resolution BUT YOU MUST band limit at say 22k for every processing you do. Soo.... Oversampling. But sometimes plugins don't have oversampling. Alot of times actually. That's why you wanna run high samplerate. AND band limit the signal

[DJ Warmonger]

I just run everything at 88200 Hz. This is actually the only way to avoid aliasing - to filter stuff beyond 22050 Hz range the signal needs to be over LP filter frequency in first plac,m which is not possible with default 44100 Hz sampling rate.

And yes, the difference in clarity and high range is audible.

[ampetrosillo]

Not quite. You should use a high resolution BUT YOU MUST band limit at say 22k for every processing you do. Soo.... Oversampling. But sometimes plugins don't have oversampling. Alot of times actually. That's why you wanna run high samplerate. AND band limit the signal

If oversampling then employs an IIR brickwall filter, then you are using a minimum-phase (latency-free) filter. Otherwise, and as it happens, in most cases, it is a FIR linear-phase filter. If you run plugins at 88.2kHz or 96kHz natively you have the possibility to use minimum-phase filters after non-linearities when you don't need phase linearity (which is very often; you don't need to preserve phase relationships between different instruments, generally). That means you'll only get regular ringing which does not impact transients as much as pre-ringing, at least academically speaking.

[Michey]

No!

No

Yes.

This had me bursting out in a loud laugh

Why ? it reminds me of the legendary Monty Python sketch called "Nudge Nudge" (jump straight to 01:56 Min):





For me, there no comparison between 44.1 and 96 as far as PROCESSING goes. and although 96 does not eradicate aliasing altogether (read this), in can further refine the process.
44.1 with or without OS is just no go.
At 96 everything PROCESSING WISE feels just "right". the bass response, the highs are not coarse, the ease of manipulating exactly what you want to manipulate (eg. EQing)

[Modular Manfred]

Yes.

I've been doing this and advocating it for years. If you can work at 96k and it sounds better at 96k then use 96k. Some plugins dont work properly at anything but 44.1k because the dev didnt code properly in which case there may be no way to work at a higher res. When everything works in 96k, I mix and master in 44.1 and render to 96k and it often sounds superior to lower rates.

In my experience the golden way is to work on your preferred or final sample rate and only use plugins that internally oversample. Without oversampling you won’t rid off aliasing and if you use analogue modelled plugins then this is crucial. Best of both worlds but you need to be selective with your plugins.

[kPere]

As far as the discussion goes I'll keep my popcorn and keep checking it, but what I find really important is - if that actually matters? - surely in many cases the difference between 44.1 and 96 may be audible, if its changing or taking the quality to another level is one matter, but the other is the fact that you can hear the difference by A/B the same exact material, which is pretty academic approach - in real world we just work by ear, and what I wonder is if 2 mixing engineers (both profesionals) given the same material would mix it (one on 96 and 2nd one on 44.1 with ocasional oversampling) and if audience could hear which one is better quality (not artistically)
I'm not sure actually..
But trying to get as good quality as you can is a good thing nevertheless. Will keep looking at this thread since I might skip to 96 some day.

[plexuss]

As far as the discussion goes I'll keep my popcorn and keep checking it, but what I find really important is - if that actually matters? - surely in many cases the difference between 44.1 and 96 may be audible, if its changing or taking the quality to another level is one matter, but the other is the fact that you can hear the difference by A/B the same exact material, which is pretty academic approach - in real world we just work by ear, and what I wonder is if 2 mixing engineers (both profesionals) given the same material would mix it (one on 96 and 2nd one on 44.1 with ocasional oversampling) and if audience could hear which one is better quality (not artistically)
I'm not sure actually..
But trying to get as good quality as you can is a good thing nevertheless. Will keep looking at this thread since I might skip to 96 some day.

Another bit of market data that might be useful are the number of long living sites that sell "high res" music. There are a lot of music listeners that say they prefer higher data rates vs lower data rates. Are they wrong? How can you tell? Maybe you can't hear the difference even if you are "professional". Logically it makes sense to work at high rates if possible. Market-wise there some evidence that there is a market for "high res" audio. There really arent a lot of compelling reasons to work at lower rates unless file size is a constraint.