Busting Aliasing myths being propagated in Audio
Вставка
- Опубліковано 6 вер 2024
- Gaurav talks about aliasing when pushing plugins. All saturation and compressor plugins will create aliasing to some degree even when oversampled.
This is just how digital works. He provides a listening test at the end of the video for the users to actually check for themselves if they are even able to hear “ aliasing “
Check out our AI-based Tape Plugin :
tone-empire.co...
Check out LVL01 :
tone-empire.co...
#ai #tape #audioplugins
I was able to correctly guess the oversampled version as well. The oversampled version has more clarity in the highs (although quite subtle) which makes sense because aliasing foldback is creating inharmonic muddiness progressively towards nyquist.
However, oversampling can in fact mess with transients. From my understanding, this is because of phase shift/smearing from the low-pass filter applied to the signal to bring it back to the base sampling rate. There may also be artifacts from upsampling interpolation.
It's a trade-off of benefit and detriment. Although one track may not show much difference, it adds up with more and more tracks. I still prefer to have it as an option - since in certain cases the benefit outweighs the cost. I'd love to see it on your AI plugins. You are doing great work and I appreciate that you are always pushing to improve and push boundaries!
True, the difference is clear
To be really honest and objective, I perceive more definition at the upper end of the frequency range using the ear, in the A/B test. It is clear that the difference is small, but I think that what people expect from a complement that emulates analog equipment, is that the problems derived from the digital world, such as digitally generated artifacts, will be solved.
The problem is that these artifacts accumulate in the mix during processing, generating a horrible and undefined sensation at the upper end of the frequency spectrum and this is audible.
I think that a test with a mix of several instruments, processed with the same plugin, would be much more noticeable the accumulation of artifacts in the upper end of the spectrum.
In REAPER, they have implemented an Oversampling system to be able to use it with any plugin without the need to have it built-in.
Also, I sincerely think that if you want to defend something, you have to do it not only with subjective arguments, but with objective evidence to prove it.
I think that this technology is very powerful, but perhaps the oversampling is too much extra processing for the CPU and that is why it has been possible to do without it.
I bet if you did a blind A/B/X test on different material you would choose base on your preference. There are a lot of processes that can cause muddiness: compression, saturation, distortion, eq. delay, your monitors you name it. The issue now is that something sounding "wrong" is being attributed to aliasing whereas aliasing is usually a very minimal component in audio coherence. In many instances you might even prefer the aliased version!
I generally use oversampling to change the character of a plugin. That's it!
Sometimes oversampling sounds better sometime no oversampling sounds better. The important part is the "SOUNDS BETTER" part which is ALWAYS a subjective judgment. Trying to quantify "sonically good" through graphs reminds me of the old "dancing about architecture" quote.
It's weird that people keep getting hung up on things like aliasing on non linear processes. You've just smashed the audio to bits and people are worried about foldback distortion? How do they even pull them apart? lol.
It reminds me of the sample rate wars. Very few people could hear any difference but people were blindly doubling their processing and storage overheads for little or, let's face it, no difference.
Creating music has become far too visual and not using a certain plugin because its output LOOKS funny is one of the most ridiculous concepts in existence.
👏 perfectly stated :)
That's insane! Even over UA-cam the non-oversampled audio clearly has more transients audible in the ride cymbal
😅 see ! 😊
I've spent a while trying to overcome all of my aliasing worries. I did find a solution, but ended up going back to mixing without antialiasing. Recently, this has popped back up, since the insurgence of Machine Learning, AI, Neural Networks/ IR samples and convolution. I've learned that aliasing CAN in fact make a mix less detailed in the high end. My old remedy that still works is a plugin from Airwindows, called Hypersonic. By cutting all the frequencies above 20k, prior to saturation, the mix is indeed much more clear. I use oversampling when the option is there, as well. Thanks for chiming in on this issue , it's one that many face and discussion can give clarity to those who are questioning their hearing.
A more interesting audio test would be piano - a multi tonal instrument, and flute, a single tonal instrument with harmonics. Drums are a tricky thing to rate - yes, they are different but so what - is one better or worse, that depends on what you want. One of the problems with aliasing in plugins is that it can add up in slightly annoying ways that a single plugin alone wouldn’t matter or notice.
So true as well :)
Again, I appreciate you tackling topics related to your work with openness and even a generous attitude toward those who might disagree with you. Class act.
All that matters is the sound, and I like the sound of your work. Lochness hooked me, and I look forward to your future products.
🙌
the difference is clear, the version with the aliasing is likely the same but with a "soft plaid" on the highs, less air.
the Aliasing is inaudible itself but the effect of it is clearly audible.
That’s the transient response change that is clearly audible , aliasing is v v minutely evident ..
Yep. I even hear it when it's on the drums on one track. It's subtle but I hear it. Not so subtle as it starts infiltrating your entire mix
What I don't understand is why there's even a debate/discussion. It's a known audible flaw in digital audio with an available solution. It makes sense that if it's technically possible, then all plugins which create harmonics should have an oversampling or filtering option - that way the end user has full creative control... which is the point after all.
Foldback aliasing will create inharmonic noise, which can be either a color in your color palette or something to minimize if you want a clean sound.
Giving the option to control this aliasing with oversampling is important.
I don't want to show off, I'm completely honest here: I listened to this on my smartphone (!) And it turned out that it was the oversampled version I liked better. The oversampled compression sounded more open to me while the track without OS sounded somehow crowded and smeared. I'm definitely no audio snob, I really wanted to feel confirmed that OS isn't that big of a deal. Apparently it is. Perhaps it is because I've been doing audio engineering for 23 years and I know what to listen for. 🤷
Probably the typical consumer of our mixes wouldn't be able to tell a difference. The thing is: They are not the ones who shall get attracted, are they? We (audio engineers) are the ones who are supposed to buy your plugins.
BTW, got Goliath and I really like its character!
The oversampled version definitely sounds smoother and more pleasant to my ears. Listening on my LCD-X headphones.
Sure , but do you hear aliasing ? .. it’s a whistling pitched sound . The difference you’re hearing is change in transient response .
@@gdayal9 well I think aliasing is any frequency beyond niquist that is folded back into the audio as a random harmonic. It doesn't fit musically. To me it just sounds dirtier than the oversampled signal but not in a good way. BTW I absolutely LOVE the tape plugin so please take all this with a pinch of salt :) You're doing great stuff man!
@@edmcguire4926 I'm totally cool, I can be wrong, I appreciate your kind response, thank you for your input :) Thank you also for being a customer ! 🙏
at the moment oversampling in AI models doesn't improve aliasing , but sampling datasets at high rates such as 192K has helped :) TM700 has that :)
@@gdayal9 the tape plugin is sounding amazing thank you! Would you ever consider doing a video where you went over how you developed a plugin like this one?
I can hear aliasing. And even though it is harder to detect when it's on one track in your mix.... As it builds up it does the opposite of what could unlock your does. .. It gets sweeter as it builds up.... Alison gets unmusical and create this brittle crust on top of the mix. When it's gone the music feels really different.
Just listening on my broken iphone speaker I preferred “B”. The transients on”A” sounded dirtier. “B” sounded cleaner. I made my decision before the results were made at the end. Now the real test is to do this with individual tracks over an entire tune and see how that “ stacks up”.
Thank you for taking on and demonstrating your take on this controversial topic Gaurav!
Had to do it sir ! Hearing is the best test :)
the oversampled one was obvious to me. it is clearer, more defined overall, and less confused/harsh at the high end. but yes OS can introduce possible transient and phase (for non-linear phase OS) changes. also some OS' AA filters roll off treble noticeably, and then the dangers is that it sounds 'smoother' when actually you may just be hearing that high treble rolloff.
if you've been mixing without OS, you do need to pay attention to (and possibly compensate for) any unwanted tonal changes when you enable it.
My guess was oversampling was B all the time, Much cleaner highs. I use oversampling in de-essers and fabfilter pro c-2 a lot. Sometimes the oversampling takes away the punch but not if it's done right.
I guessed correctly, but I'll admit that I usually find it hard to hear aliasing(maybe I got lucky). In this example I thought the oversampled drums sounded slightly more punchy and open. They felt more forward while the non-oversampled drums felt ever so slightly restrained.
I didn't think that it would make a different. Seeing as I already purchased all of Tone Empire plugins, Even the recently released TM700. Basically, I support the company. With that said, I assume this was a trick and didn't think it was going to be the same audio played to compare against each other. Unfortunately there was a clear difference. A, sound like it had this light grainy sound in the air. B, was smooth and you can hear the space without the grain. Whatever that means, I hope you do offer the oversample on every product you offer, because that is two different flavors that would enhance the versatility of the plugins. Whether is a myth or not, the usability of the plugins can be increase with the oversampled option. This doesn't effect me anyway, as I've already purchased all your plugins, but as a customer I would appreciate all the options to make the tools you offer more appealing. Thanks again for this video.
The day oversampling Help’s AI based models , or whatever process helps improve quality , it will be added . As of now it doesn’t . That’s the point as well . Feel free to use oversampling :) that’s your choice where available .
You are correct G. I often choose to not use so much os on my master buss because when I A/B I can clearly hear the loss in transients. I know I personally wrote support about this. I don't think the issue is if the aliasing exists or not, at least I know it's not with me. I would just like the choice! I would like to have oversampling as an option an any compressor or saturation tool I use. Sometimes there is a benefit and sometimes there isn't but I would like to be able to make that choice myself. That was my reason for writing an email to support. It was a feature request. In my personal opinion, I think os should be an option on any plugin that has saturation or compression. I still bought both lvl 1 and tm700. I bought them to support the company. I will still use them both sometimes but I would use them a lot more if I had the option to os the plugins. 🙂
Cool. Interesting subject. Also another "must watch" video about aliasing is by Dan Worrall. called: Linear, Logarithmic, Exponential & Perspective
Correct , seen that one , great video ! :)
@@gdayal9 I know right? ! Its like everyone is on "error hunt" and even though we have the best tools ever made , be that analog or digital , there is a ton of boring music being produced with them. Not saying all music is trash but I think you get my point.
Same as synths. Most great musicians I know don't complain about their instruments and their capabilities all the time. It is mainly the boring music making people who complain the most. Paradox isn't it :-)
Watching videos of sessions of top levels professionals, they clearly don't care about aliasing in their analog emulations. Because they rarely manifest into their sound. They use many instances of plugins across their mix which you cannot do with a complete antialiased plugins due to low headroom on CPU. It's the lesser known producers who complain too much about aliasing. I think people are over-thinking the over-sampling too much. Sometimes I do turn on oversampling at maximum on high gain amp sims and render the guitar track because those might be the rare cases in which aliasing are noticeable due to the nature of extreme distortion. But honestly that's only exception.
True and Well Stated !
I could hear a small difference, I guessed which one was oversampled before you revealed it, because it sounded very slightly cleaner (listen to the ghost notes on the snare they sound a tiny bit crunchy on the audio that is not oversampled). I doubt in a full mix you would notice any difference though. Maybe I am hearing the oversampling messing with the transient and not Aliasing though
Correct it’s the transient response change , not aliasing
@@gdayal9 I actually preferred the version that was not Oversampled, slightly more Vibey to my ears
First off, I definitely hear the difference and prefer the oversampled version, but I’m pretty sure it’s just because of the transient smoothing, like you said. However, that ultimately lays a claim for oversampling in plugins regardless of what it’s doing for aliasing. I’ve done a lot of aliasing tests, and I, like you, think it's generally getting a bit out of hand. I recreated the test White Sea did involving the LVL01, and what he leaves out about the aliasing is that at its loudest volume, it’s only registering around -80db compared to the sweep signal, which is at -20db. In a real-world scenario, that’s going to be inaudible. For reference, it’s approximately the same results you’d get using the UAD 1176 at a heavily saturated setting. I believe UAD uses internal oversampling to taste when they feel it’s necessary. I actually just recently emailed you guys about this and suggested you do a video about it, so I’d like to think I had a hand in getting this video made. Lol, but I don't know. Either way, the point of all this is obviously to use your ears. But I also think that since upsampling can result in better sound, then why not make it an option when creating these types of plugins? I’d suggest someone create an easy-to-use, free plugin skin that will upsample any plugin for you. Then it’s a win/win.
Oooor....you could use Reaper as your DAW which is able to natively oversample the whole FX chain or single plugins in your chain 😊
@@Mansardian yeah, I've heard that Reaper can do that. That is pretty awesome. definitely a reason to try out Reaper.
I think I heard a little of that transient smearing on version B, so I guessed correctly that it is the version with oversampling. I actually prefered version A.
The model 5000 sounds great on rock guitars too.
Would love to hear the effects on something a bit more
Tonal like strings or woodwind solo. Would love to try this on some more exposed tonal elements and like you say … on a few tracks I suppose there is an element o It adding up! I wonder if any of you engineers cringe as much as some of the slightly more Controversial UA-camrs trying to get views at the sound of any aliasing though especially
When you have clients in? I think the music and feel is far more important realistically. Interesting stuff. 😂
I’d like to know more about why the oversampling doesn’t work well on the Ai Models and how exactly it causes the transient issues particularly as more and more AI created plugins will appear soon.
Preferred the non-oversampled! Felt clearer and more "open" to me, I guess due to the difference in transient response?
I'm always confused about oversampling, I know it's better for aliasing but for me i always don't like the effect to low end when oversampling on..Thanks Gaurav ❤
Aliasing is not a big deal on drums as it's not a tonal instrument, it just adds noise to noise. On tonal instruments, aliasing will create out of tone pitch artefacts that anybody can hear.
Not always , depends on how much one is pushing a system .. but most confuse distortion with aliasing .. and yes sure it’s more evident in certain material :)
@@gdayal9 it also creates a certain cloudiness that certainly does accumulate over a mix. If you compare a heavily oversampled mix with lots of saturation, with a non oversampled one- an audible difference can be heard. The OS one can be perceived as more open and having increased clarity.
Very interesting to hear about OS in relation to compression specifically
@@Rhuggins sure :)
Personally I do think it should be up to the developer to decide whether oversampling is necessary or not for the plugin to sound/work better. All the oversampling, auto gain, etc hype on the internet has gotten a little out of hand and I do think people really lose sight of the art and forget to use their ears sometimes. On top of that using oversampling isn't always the best solution as you mentioned especially with certain transient material. I really appreciate when developers do oversampling under the hood if necessary because to me it's not really a creative choice a mixer should be making. It's like we have to choose how "well" the plugin works at the expense of CPU load but in my opinion that should be a choice the developer makes when considering the efficiency and intended sound of the plugin. With all that being said though I thought the oversampled x4 version sounded better in this example 😅
🙏
Not only sometimes. Many simply replicate what they see in videos like "look this plugin response line is 3mm off ! How dare thy!!" but most times without really (really) hearing a difference
the comments on aliasing always make me turn the video off from the other folks. also when they just set the knobs wrong and say it doesnt compress enough it is a bit frustrating. they are acting like they are saving everyone from a scam, but are also acting a bit scammy themselves. painters buy paint mixers buy plugins. yours are good for certain things and they are a good price. I wish the audio UA-cam scene was dominated by more useful and constructive things rather than about analog or digital. I like the new tape plug, and lvl sounds good on a mix test though I haven't really used it yet.
I was just thinking about aliasing, and the sine sweep reveals the sound. but what happens when the signal is more complex? wouldn't the sound of the aliasing overlap itself so much that it no longer has a pitch and is then just low level noise? much like that noise in any audio signal chain?
could u create a video that shows that sound? on some level I get why aliasing is happening, but it is also confusing lol
Exactly :)
It's most evident in the top end? How is your hearing for high frequency content?
I heard a difference between the two, with B sounding a little brighter to me, and perhaps the transient of the snare sounding as if it had more attack. Can’t say if I’d prefer one over the other, would need to be in context of a mix. I do think that we have really gone down the rabbit hole of analyzing plugins to death. I guess I’m in the camp of if it sounds good it is good. If it translates and people enjoy it then you’ve done your job, learn from the experience and carry it on into the next mix. I will say I’ve demoed the new tape plug-in, have you considered adding some VU meters to it so we can get a sense of just how hard we are hitting it?
I could tell the difference between oversampled and not while listening to this over UA-cam on my phone speakers
first of all, the entire loop sounds distorted, in all instances, so that might be a source of confusion. other than that, to my ears, a sounds a little more bloated, and the brilliance comes out a little better on b. actually to put it better, both have a little mud, but for b most of it seems to move a little higher in the low mids, which makes it sound a little more balanced overall. so i would go with b.
Surprisingly, B sounds a bit thinner... 😮
:)
For me, this video clearly proves beyond a doubt that aliasing impacts the sound of the plugin. I can certainly hear a difference between A and B. I have to ask, you really don't hear that B is more clear, lively and open?
If it is its so marginal I kept second guessing it. When you're down to a margin that thin you have to ask whether the gains are worth the performance hit. I listened to this for like 4 or 5 minutes and I can't hear it at all when I turn my screen off and try to guess. At first I was sure B at first sounded a tiny bit brighter but then I realised it was because the very first ride hit is brighter / louder than the rest and I think that coming in on the first switch to be altered my perception. I certainly wouldn't go anywhere near suggesting there is a significant difference beyond a shadow of a doubt
Yes but that is also due to the change in transient response , aliasing might be a tiny tiny factor
@@gdayal9 What is causing the change in transient response in your view and why does this change result in B sounding more open that A?
@@ScottMcFadyen it’s grabbing less peaks .. it’s unintentional , it’s should be compressing
The “clarity” being perceived is due to transients passing through .. this is not intended :) an uncompressed track will sound more clear than a smashed one .
Is this a response to @whiteseastudio ?
It’s just a take on being blindly influenced … if one wants to show something in a bad light , it’s not hard to do that :)
very likely.There wouldn't be a reason for a plugin developer to make such a video otherwise.
You may mistake aliasing with distortion… but my friend, aliasing IS (a type of) distortion.
Absolutely correct :) I should've added a demo of what aliasing actually does sound like, so as to make reference to the type of distortion :)
Too much distortion also spent sound right.... It was something doesn't mess with the transients if you lose linear phase and to your ultrasonic filters right.... It really is not something that people are going to hear when done right. Aliasing when it starts getting in every track it builds up and it totally doesn't sound right.
Can the average person who buys or streams your audio hear aliasing? Nope. Andrew Sheps mixes in the box with plugins that what?…. You guessed it
Yes ! All music from the 90s - 2015 should be rejected then if aliasing is a thing to consider ;)
sir thanks english tutorial bhut hai hum app ka hindi me tutorial banye thanks