You got close, but not quite there. First up, yes, good point on using oversampling to "catch" harmonics created by processing, this is a good point. I think the phenomenon of aliasing should be explained some more though. To say it in the simplest form I can think of: at 44.1kHz, the limit (Nyquist) frequency is half of it, 22.05kHz. If you try to create, record or process anything above that, the frequency "bounces back". If you try to record or create 23kHz sine wave, it jumps and shows up at 21kHz. Now, it wouldn't be a problem normally (we have harmonic distortion in all of the gear anyway), if it wasn't for one really important detail. After the frequencies "bounce", they are no longer "harmonics". They aren't related to the original signal and all because they are not a multiplier of the fundamental frequency. This means we get inharmonic sound. This results in a buzzy and nasty sound. Well that sounds good, let's just not create anything up there, right? It's not so easy, though. High frequency equals fast jump in the waveform. A perfect square wave would have an unlimited number of harmonics, which (you guessed it) jump back into the audible spectrum so together with a note played on a virtual synthesizer we also get a random buzzing tone. When creating a plugin it needs to be taken care of. You can't just put a lowpass filter on top of it, because the frequencies aren't above - there isn't anything above 22.05kHz in terms of digital audio. You'd have to pick and choose specific harmonics, which gets messy because it's essentially glued together with the sound we're trying to create. You'd say "but I don't use synthesizers, what's the point". Well, if you do distortion, "saturation", "tape emulation" or whatever gimmick name it's called this week, you do create harmonics above the input, don't you? If a single one of them jumps above 22.05kHz it gets shoved down and buzzes together with your input. Same with compression and limiting. If you don't know, distortion equals compressor with no attack and no release. The whole point of a compressor is to add the time constants to make it more subtle than obvious clipping. And what's the result of distortion (= compression)? Square wave. Ever so subtle, but still. Fast jump in waveform. Which equals high frequency. If you don't keep more sample points to catch it, it bounces back into the input.
I believe it's possible to design a filter that can be integrated into the limiter that can remove the frequencies above 22.05kHz before they bounce back. Maybe not done yet, but quite possible.
So what is the point of oversampling in a cd player? Unless I am not understanding this at all, this is after the fact. You cannot get data off the CD that doesn't exist. Even if you "re-sample" the stream into 88.2khz, you are dealing with a stream that was recorded at 44.1. So what does it do and is it really necessary?
Okay, it's time for me to check for some spare cash the month to become a patron. I love you doing more of these "technical" videos and while I have a very basic understand about the most necessary DSP basics and a little beyond that, I lack knowledge of what goes "beyond". Great video, keep em coming! :)
Oversampling became a very important part for me when choosing plugins, hence why I love TDR plugins so much. This also applies to ReaComp that has even up to x64 of oversampling (I tend to not use higher than x8), and when it comes to analog-modeling plugins those usually already have oversampling, they just don't show it but the manual usually specifies it. I gues aliasing along with the noise floor (the one fixed with dither) are the two main problems why some engineers in the past (and some still today) say that analog sounds better, because those two make the sound worse in many ways.
6 лет назад+1
Understood in theory, now I need practice. Thanks for all your videos, I love them
Light waves also come in many frequencies. The frequency is the number of waves that pass a point in space during any time interval, usually one second. We measure it in units of cycles (waves) per second, or hertz. The frequency of visible light is referred to as color, and ranges from 430 trillion hertz, seen as red, to 750 trillion hertz, seen as violet. Again, the full range of frequencies extends beyond the visible portion, from less than 3 billion hertz, as in radio waves, to greater than 3 billion billion hertz (3 x 1019), as in gamma rays.
I demo'ed Fabfilter Pro L2/ C2/ Q2 Plugins, on my tracks which were recorded at 96 kHz, 24 bit, and tended to hear that my audio had lost slight clarity, like it was covered in a thin, plastic 'veil'. I read somewhere that Fabfilter plugs don't like high sample rates like 96 Khz. Is there an ideal setting in oversampling to compensate for this? I tried 8x oversampling, but from memory the problem still existed.
What about RMS limiting on high treble? Doesn't a little oversampling make the RMS measurement of high treble more accurate / less dependent on the sampling phase? Don't know whether the difference would noticeable without a pathological signal.
i really like the sound of 48khz aliasing and i wish somebody would make a plugin to intentionally alias (yes i know there are bit crushers but most of them don't give you access to sample rates above 22k)
wait im still a bit confused. should i change all my vsts to oversampling or just master bus vsts or mainly when using a compressor/ and or limiter with a fast attack.
Generally: use oversampling when working at 48k or lower. Working at high rates, it's less important. More important than following rules, though, is testing things and using your ears.
If your system can't handle oversmapling all the plugins that allow it, what you can do (if you have offline bouncing like most DAWs these days) is mix with the highest oversampling possible within your systems limitations and then when you bounce at the maximum settings. Some plugins allow you to set them so they oversample during bounce but can be set to not do it during normal operation.
What you should do is at least watch your analyzer when using high frequencies. If you notice the spectrum band stays high on the top-end, filter out the top-frequencies. At least just below its Nyquist-limt. Using higher samplerates does not take away aliasing issues, it just makes you more aware of what you are recording, in this case high frequencies. Especially stuff like noise/ hiss/ high pitched freq's that you didnt hear in the first place, but pop up on the analyzer.
thanks for the video. very usable information. but one question, isnt 32 times oversampling not to much ? what is the benefit of such a high oversampling ?
This may be contrary to what most people believe, but I typically don't like the sound of oversampling. Different software companies use different filters for oversampling, but some (fabfilter comes to mind immediately) smear transients in a way that is not pleasing at all to me. I don't notice it in all implementations of oversampling, but I always compare non-oversampled to oversampled to see if I like one or the other better. Plugins that create a lot of harmonic content, I tend to prefer oversampling. But for most limiters, I tend to turn it off because it keeps the transients more intact.
So fabfilter plugins can oversample within the plugin, but they’re so expensive! Is there another way to oversample and filter out those high frequencies that will cause aliasing?
I don't see what's complicated about this. You just up the rez. In horizontal direction. On second thought, some simple visualization would have gone a long way.
Хобби
I watched this with headache, now i unironicly have no headache anymore
You got close, but not quite there. First up, yes, good point on using oversampling to "catch" harmonics created by processing, this is a good point. I think the phenomenon of aliasing should be explained some more though. To say it in the simplest form I can think of: at 44.1kHz, the limit (Nyquist) frequency is half of it, 22.05kHz. If you try to create, record or process anything above that, the frequency "bounces back". If you try to record or create 23kHz sine wave, it jumps and shows up at 21kHz. Now, it wouldn't be a problem normally (we have harmonic distortion in all of the gear anyway), if it wasn't for one really important detail. After the frequencies "bounce", they are no longer "harmonics". They aren't related to the original signal and all because they are not a multiplier of the fundamental frequency. This means we get inharmonic sound. This results in a buzzy and nasty sound. Well that sounds good, let's just not create anything up there, right? It's not so easy, though. High frequency equals fast jump in the waveform. A perfect square wave would have an unlimited number of harmonics, which (you guessed it) jump back into the audible spectrum so together with a note played on a virtual synthesizer we also get a random buzzing tone. When creating a plugin it needs to be taken care of. You can't just put a lowpass filter on top of it, because the frequencies aren't above - there isn't anything above 22.05kHz in terms of digital audio. You'd have to pick and choose specific harmonics, which gets messy because it's essentially glued together with the sound we're trying to create. You'd say "but I don't use synthesizers, what's the point". Well, if you do distortion, "saturation", "tape emulation" or whatever gimmick name it's called this week, you do create harmonics above the input, don't you? If a single one of them jumps above 22.05kHz it gets shoved down and buzzes together with your input. Same with compression and limiting. If you don't know, distortion equals compressor with no attack and no release. The whole point of a compressor is to add the time constants to make it more subtle than obvious clipping. And what's the result of distortion (= compression)? Square wave. Ever so subtle, but still. Fast jump in waveform. Which equals high frequency. If you don't keep more sample points to catch it, it bounces back into the input.
teach me more
I believe it's possible to design a filter that can be integrated into the limiter that can remove the frequencies above 22.05kHz before they bounce back. Maybe not done yet, but quite possible.
Thanks! That helps a lot in clearing the doubts. To avoid all these we use oversampling!?
Agreed, it's way easier to explain or visualize...
So what is the point of oversampling in a cd player?
Unless I am not understanding this at all, this is after the fact. You cannot get data off the CD that doesn't exist. Even if you "re-sample" the stream into 88.2khz, you are dealing with a stream that was recorded at 44.1.
So what does it do and is it really necessary?
Okay, it's time for me to check for some spare cash the month to become a patron. I love you doing more of these "technical" videos and while I have a very basic understand about the most necessary DSP basics and a little beyond that, I lack knowledge of what goes "beyond". Great video, keep em coming! :)
Oversampling became a very important part for me when choosing plugins, hence why I love TDR plugins so much. This also applies to ReaComp that has even up to x64 of oversampling (I tend to not use higher than x8), and when it comes to analog-modeling plugins those usually already have oversampling, they just don't show it but the manual usually specifies it.
I gues aliasing along with the noise floor (the one fixed with dither) are the two main problems why some engineers in the past (and some still today) say that analog sounds better, because those two make the sound worse in many ways.
Understood in theory, now I need practice. Thanks for all your videos, I love them
Light waves also come in many frequencies. The frequency is the number of waves that pass a point in space during any time interval, usually one second. We measure it in units of cycles (waves) per second, or hertz. The frequency of visible light is referred to as color, and ranges from 430 trillion hertz, seen as red, to 750 trillion hertz, seen as violet. Again, the full range of frequencies extends beyond the visible portion, from less than 3 billion hertz, as in radio waves, to greater than 3 billion billion hertz (3 x 1019), as in gamma rays.
Thank you. Well done and I think another video on the subject would be cool too.
Great explanation. Thanks!
I demo'ed Fabfilter Pro L2/ C2/ Q2 Plugins, on my tracks which were recorded at 96 kHz, 24 bit, and tended to hear that my audio had lost slight clarity, like it was covered in a thin, plastic 'veil'. I read somewhere that Fabfilter plugs don't like high sample rates like 96 Khz. Is there an ideal setting in oversampling to compensate for this? I tried 8x oversampling, but from memory the problem still existed.
Hi, what oversampling rate should we use in pro q 3? I aim for something professional but not do anything overkill that is unnecessary.
Now, understood! Thanks
As a new comer to this subject, thank you! Great job.
Leuk dat er ook iemand is die weet waar hij over praat. Leuke video man ga zo door!
What about RMS limiting on high treble? Doesn't a little oversampling make the RMS measurement of high treble more accurate / less dependent on the sampling phase? Don't know whether the difference would noticeable without a pathological signal.
Great video! You're really good at explaining this stuff, looking forward to more video's like this.
Brilliantly explained!
So how are those high freq’s related to us if we can only hear untill 20KHz?
How does oversampling apply to video... for example oversampling 4k video or 6k video
i really like the sound of 48khz aliasing and i wish somebody would make a plugin to intentionally alias (yes i know there are bit crushers but most of them don't give you access to sample rates above 22k)
That’s a great idea! On some genre’s, I love the aliassing effect!
Check out Airwindows, he did it...
wait im still a bit confused. should i change all my vsts to oversampling or just master bus vsts or mainly when using a compressor/ and or limiter with a fast attack.
You can just upsample the source files and mix them down, and then downsample the result.
Generally: use oversampling when working at 48k or lower. Working at high rates, it's less important.
More important than following rules, though, is testing things and using your ears.
If your system can't handle oversmapling all the plugins that allow it, what you can do (if you have offline bouncing like most DAWs these days) is mix with the highest oversampling possible within your systems limitations and then when you bounce at the maximum settings. Some plugins allow you to set them so they oversample during bounce but can be set to not do it during normal operation.
What you should do is at least watch your analyzer when using high frequencies. If you notice the spectrum band stays high on the top-end, filter out the top-frequencies. At least just below its Nyquist-limt.
Using higher samplerates does not take away aliasing issues, it just makes you more aware of what you are recording, in this case high frequencies. Especially stuff like noise/ hiss/ high pitched freq's that you didnt hear in the first place, but pop up on the analyzer.
thanks for the video. very usable information. but one question, isnt 32 times oversampling not to much ? what is the benefit of such a high oversampling ?
That is a question that I also still don't know the answer to... I think 32x is a simple matter of: Why? Because we can!
Good video man
This may be contrary to what most people believe, but I typically don't like the sound of oversampling. Different software companies use different filters for oversampling, but some (fabfilter comes to mind immediately) smear transients in a way that is not pleasing at all to me.
I don't notice it in all implementations of oversampling, but I always compare non-oversampled to oversampled to see if I like one or the other better.
Plugins that create a lot of harmonic content, I tend to prefer oversampling. But for most limiters, I tend to turn it off because it keeps the transients more intact.
So fabfilter plugins can oversample within the plugin, but they’re so expensive! Is there another way to oversample and filter out those high frequencies that will cause aliasing?
thanks man
I dont get it
Hi, check this: Voxengo OVC-128, 128X linear-phase oversampling. why? 😭
I don't see what's complicated about this. You just up the rez. In horizontal direction.
On second thought, some simple visualization would have gone a long way.
Makes sense
So 48kHz ist not an option?
thanx. would've been cool to see it after it was oversampled! but thanx for the rocket science ✌😁
Make a new video on the topic please
starts at 03:02
you lost me at the 1:30 mark XDDD
Haha you said shit.
88200 👍
you are the skrillex of youtube my friend
With other words computer recording is still shit! Nothing better as recording on tape why its called a real to real!
I think it’s funny and it’s hard to take all the technical stuff you say seriously, when you drop the word “shit” loquaciously.