OBS Audio Monitoring

[Gammbet]

Hello!

I would appreciate some insight about the following:

For what I understand, OBS maximum Audio Sample rate is 48khz.(for streaming / recording)
If I'm using an imput device running 96khz and I turn on Audio Monitoring on OBS for that device, will I output 48khz or 96khz?

Setting -> Audio - > Advance -> Monitoring Device

 

[AaronD]

Probably 48k. And it might be unusable because someone chose to write the Audio Monitor so that it locks to that device and adjusts the buffer as needed *instead of resampling*! I have no idea in what universe that's useful, but that's what they did.

Any amount of clock mismatch at all will either cause underruns or the buffer will grow to the point of being uselessly delayed. How quickly that happens depends on the amount of mismatch. For nominally the same setting but from a different physical clock, it might take a while; but for different settings, it happens pretty quick. If *Windows* resamples for you, then that can save your bacon, but OBS is still providing its internal sample rate to that external resampler.

---

That said, I really don't see a benefit to anything beyond 48k. Contrary to audiophool belief, it doesn't sound *any* different *at all*. Nyquist (the highest frequency that can be represented accurately) for 48kHz is 24kHz, which is already beyond human hearing. And no, we don't perceive harmonics of that upper limit.

For timing and other things, see here:

Xiph.Org Video Presentations: Digital Show & Tell

www.xiph.org

---

With THAT said, I *am* designing a bare-metal project (no operating system at all) to use 96kHz audio (or possibly even faster), not for the non-existent sound quality, but to allow the analog <-> digital converter chips themselves to use less aggressive lowpass filters, which in turn provide fewer samples of latency in addition to having less time between samples. This is for a mic and speaker physically close to each other, and the entire system has to line up with the speed of sound over that short distance. Again, it has nothing to do with the sound quality!

OBS has enough else going on, not to mention a multitasking operating system (Windows, Mac, Linux, etc.), that it needs to have a buffer that can fill up while it does everything else. Then it processes that buffer, and lets it play out at the end while it does something else again. So it's completely impossible to realize that benefit, and thus no point in running the converters that fast.

No need for anything beyond 48kHz if you're using enough of a buffer to make a multitasking system work.

 

[Gammbet]

Probably 48k. And it might be unusable because someone chose to write the Audio Monitor so that it locks to that device and adjusts the buffer as needed *instead of resampling*! I have no idea in what universe that's useful, but that's what they did.

Any amount of clock mismatch at all will either cause underruns or the buffer will grow to the point of being uselessly delayed. How quickly that happens depends on the amount of mismatch. For nominally the same setting but from a different physical clock, it might take a while; but for different settings, it happens pretty quick. If *Windows* resamples for you, then that can save your bacon, but OBS is still providing its internal sample rate to that external resampler.

---

That said, I really don't see a benefit to anything beyond 48k. Contrary to audiophool belief, it doesn't sound *any* different *at all*. Nyquist (the highest frequency that can be represented accurately) for 48kHz is 24kHz, which is already beyond human hearing. And no, we don't perceive harmonics of that upper limit.

For timing and other things, see here:

Xiph.Org Video Presentations: Digital Show & Tell

www.xiph.org

---

With THAT said, I *am* designing a bare-metal project (no operating system at all) to use 96kHz audio (or possibly even faster), not for the non-existent sound quality, but to allow the analog <-> digital converter chips themselves to use less aggressive lowpass filters, which in turn provide fewer samples of latency in addition to having less time between samples. This is for a mic and speaker physically close to each other, and the entire system has to line up with the speed of sound over that short distance. Again, it has nothing to do with the sound quality!

OBS has enough else going on, not to mention a multitasking operating system (Windows, Mac, Linux, etc.), that it needs to have a buffer that can fill up while it does everything else. Then it processes that buffer, and lets it play out at the end while it does something else again. So it's completely impossible to realize that benefit, and thus no point in running the converters that fast.

No need for anything beyond 48kHz if you're using enough of a buffer to make a multitasking system work.

I really appreciate all the information.
Thanks for taking the time to reply in such a detailed manner.

I was asking because, Im trying to sound the same on Discord as I do on Streams (I've VTS Equalizer plugin).
So what I did was, installed Virtual Cable and I'm using the Audio Monitoring feature to output the Mic to Discord with the equalisation applied.
My mic is setup at 96khz and I was wondering if I could configure Virtual Cable to 96khz.
For what I've read, it's better to set the output to 48khz due to the fact that OBS already down samples the audio, correct?

Thanks!

 

[ciddyguy]

AsronD, I think there is a misconception/misperception about audio and higher sampling in general and most of that has to do with the supposedly wider bandwidth frequency spread, which, while there is technically speaking some truth to that, much of that you may not notice much if at all anyway over the traditional 44.1 CD or even 48K.

Going to 96, or even 196K does reduce jittering, thus the noise floor (the same for using 24bits instead of 16 Bits). Thus I do notice a slight improvement in the overall noise floor (also known as dithering), the music or other audio will generally have less background noise (will sound cleaner and clearer), so you hear more of the minute details in a given piece is where the improvements mostly reside and less about frequency. You might get better detail in the lower registers, which will give you a perceived difference in the bass region, but that is largely it though.

 

[AaronD]

AsronD, I think there is a misconception/misperception about audio and higher sampling in general and most of that has to do with the supposedly wider bandwidth frequency spread, which, while there is technically speaking some truth to that, much of that you may not notice much if at all anyway over the traditional 44.1 CD or even 48K.

Going to 96, or even 196K does reduce jittering, thus the noise floor (the same for using 24bits instead of 16 Bits). Thus I do notice a slight improvement in the overall noise floor (also known as dithering), the music or other audio will generally have less background noise (will sound cleaner and clearer), so you hear more of the minute details in a given piece is where the improvements mostly reside and less about frequency. You might get better detail in the lower registers, which will give you a perceived difference in the bass region, but that is largely it though.

It doesn't even do that. Watch the video if you haven't already. Especially the demonstration of sub-sample timing.

As for a perceived difference, it takes a lot of effort that most people either don't do or don't even know TO do, to make a good test. What you need is called "double-blind", where *nobody* knows which sample is which, not even the test administrator so that they can't give subconscious clues. Write down what you think each sample is, then compare afterwards to what they really were. If you got about 50% right and 50% wrong, then you're just guessing and there's really no perceptible difference at all.

 

[ciddyguy]

I have seen that clip before and one or two others with this same guy if I recall, and I think he did one on the good ol' analogue tape on its own, but he touched on it here too.

To be fair, it's been several years since I saw that clip, but he does talk about dithering and the noise floor that I was trying to get at in my initial reply. Mind you, the changes are subtle, but are noticeable, if you know what to listen for. You are correct in that most folks do not make the effort, let alone know HOW to listen for this. I can be a tad geeky at times, and dealing with less than stellar hearing these days, but I've learned over time to know what to listen for so that makes a difference.

 

[AaronD]

To be fair, it's been several years since I saw that clip, but he does talk about dithering and the noise floor that I was trying to get at in my initial reply. Mind you, the changes are subtle, but are noticeable, if you know what to listen for.

So it's entirely about the noise floor, and the dithering trick to reduce it below 1 LSb? I can see that.

Just like everything, dithering is also constrained to Nyquist or below. If the audio range covers all of that, then there's nowhere to "move the noise to" that is not also audible. So we can only move it to where we're less sensitive, but it's still technically perceptible if you listen more closely. The purpose of a high sample rate in this context, is to give somewhere to "move the noise to" that we can't hear at all.

That said, just adding a couple more noise-free bits completely swamps this effect. Thinking about professional audio gear that is very well standardized now on 24-bit converters, and even better internal processing so that the accumulated round-off error is still less than 1 LSb at the 24-bit output. I'm pretty sure they don't dither, at least not on purpose, but the big pile o' bits means they don't need to. It's difficult but entirely possible to make an analog circuit outperform a 16-bit converter, but not even close for 24 bits. So the noise floor of a 24-bit ADC comes from the analog stuff that feeds it, not quantization, which is what dithering is supposed to fix.

By the way, I think it's interesting that people latch onto the sample rate instead of the bit depth. 32 bits per sample at 48kHz is the same bitrate as 16 bits at 96kHz. Just making the samples that much deeper completely eliminates the need to dither, for the same amount of data. And if you decide to keep your 16-bit "token" soundcard, then you probably don't care *that* much about the noise floor anyway.

Even without dither, a 16-bit 48kHz soundcard is still light-years better than a brand new cassette tape, studio tape, vinyl record, or any other analog media. Even "back in the day" when those things were actually good, and not the cheap tokens that they are now, to appeal to people who inherited the nostalgia without the memory of what it really was.

Besides, the 16-bit "token" soundcard is probably swamped with interference from the stuff close by. So dithering doesn't matter there either. For an on-board thing, your noise floor comes from the on-board switching power supplies that feed the CPU, and maybe some long-running bit patterns in a communication bus that dip all the way down to audible. You *really* must not care if you keep that one!
For the record, I do use that one in a couple of concert-capable rigs that I've set up. Considering everything else going on that I count as noise, yeah, I don't care. :-) S/N in a live setting is surprisingly low, and the mentality is very much, "close enough, just go with it". You would never get away with most of what happens live, in a recording studio.

Live broadcasts, like what most of us are doing with OBS, are somewhere in between the live and recording worlds.

 

[ciddyguy]

From the video, I got it that they just move the noise floor so low that we don't perceive it. I did notice that when you lower the lower end noise floor, the hiss dropped precipitously. When he went flat again, the hiss came back. Mind you, I was listening on a desktop through an old Technics receiver and even older mini speakers that are leagues better than computer speakers with the output of the audio being fed by the built in sound card on an older Dell Optiplex that may, or may not be 24 bit capable (I think it is?). I have an older eMachines from 2005 has has an add on card, a Soundblaster Audigy card that is capable of 24/96, tops and using Audacity, it's set to that as well (or was, the PC is now no longer usable) for audio capturing when it was in use.

Now, I mostly capture vinyl, yes, grew up on it, starting in the 70's and in 2020, finally upgraded the table to a Rega P6 but use the Grado Prestige green 1 cart with a good phono stage I built (it was from a kit) and capture at 24/96 so if I had to do any destructive editing to a file (namely to remove ticks/scratches that are bad if needed, I can keep the audio up as much as possible, even if I have to down sample to 16/44.1 for use in the car. This way, the final file is as close to 16/44.1 as possible in the end, not below it. I do find the overall quality stays good when I do it this way. Not that it wasn't good before, but it's a bit improved and I notice it. Again, the overall file is cleaner, clearer overall, but it's not a HUGE difference, unlike some like to claim.

Notice that he raises the noise floor for demonstration purposes so we can hear what he's talking about, but overall we don't hear that hiss like noise otherwise in the real world. This becomes more apparent when playing a piece that has a lot of transient details and quiet passages, like classical music or acoustic jazz pieces. The most noticeable aspect is when you go from analog to digital, the noise floor drops, which is why digital has one big advantage, you can pile on layers during the mix down sessions and not have to worry so much about the overall degradation of the final product.

 

[AaronD]

From the video, I got it that they just move the noise floor so low that we don't perceive it. I did notice that when you lower the lower end noise floor, the hiss dropped precipitously. When he went flat again, the hiss came back. Mind you, I was listening on a desktop through an old Technics receiver and even older mini speakers that are leagues better than computer speakers with the output of the audio being fed by the built in sound card on an older Dell Optiplex that may, or may not be 24 bit capable (I think it is?). I have an older eMachines from 2005 has has an add on card, a Soundblaster Audigy card that is capable of 24/96, tops and using Audacity, it's set to that as well (or was, the PC is now no longer usable) for audio capturing when it was in use.

I've also seen built-in soundcards that advertise 24/96 or even 24/192, but since they're on the same PCB as a massively noisy digital PC, I really doubt that they perform any better than the dirt-chip 16/48 ones. The conversion to analog itself might be that, but it immediately gets swamped with all of that digital noise. Same problem for input: the ADC itself might actually be that, but it always has perhaps -40dBFS of digital noise, and so the high spec is effectively worthless.

Best would be a USB sound card with a metal case, so the only digital connection it has to worry about is USB, and the faraday cage keeps everything else out. Likewise for any other cable-connected, dedicated device.

Now, I mostly capture vinyl, yes, grew up on it, starting in the 70's and in 2020, finally upgraded the table to a Rega P6 but use the Grado Prestige green 1 cart with a good phono stage I built (it was from a kit) and capture at 24/96 so if I had to do any destructive editing to a file (namely to remove ticks/scratches that are bad if needed, I can keep the audio up as much as possible, even if I have to down sample to 16/44.1 for use in the car. This way, the final file is as close to 16/44.1 as possible in the end, not below it. I do find the overall quality stays good when I do it this way. Not that it wasn't good before, but it's a bit improved and I notice it. Again, the overall file is cleaner, clearer overall, but it's not a HUGE difference, unlike some like to claim.

Yes, you've got the idea: capture 24 bits, upconvert some more to do your processing, and keep it that way until the final output, whether that's a 24-bit DAC or a 16-bit CD. But once you have that many bits, you don't need to dither, and so that reason for a high sample rate also goes away.

This thread is probably WAY more detail than @Gammbet had in mind! :-)

 

[ciddyguy]

Yeah, good points, and when I do VO stuff, I use OBS and it's filtering for audio to remove any pops and clicks that come through the computer's sound card. I don't have the budget at the moment to get a secondary recording device, such as a Zoom or similar unit so the PC will have to do, and for VO with OBS, nice to keep it all in one spot for easy of syncing as needed as it's already linked.

As for capturing audio, I don't bother with the dithering aspect, just ensure it's captured at something higher than 24/96 and when it comes to the final output, drop it back down.

As for what @Gammbit had in mind, I bet you are probably right! :-)

 

[AaronD]

I really appreciate all the information.
Thanks for taking the time to reply in such a detailed manner.

I was asking because, Im trying to sound the same on Discord as I do on Streams (I've VTS Equalizer plugin).
So what I did was, installed Virtual Cable and I'm using the Audio Monitoring feature to output the Mic to Discord with the equalisation applied.
My mic is setup at 96khz and I was wondering if I could configure Virtual Cable to 96khz.
For what I've read, it's better to set the output to 48khz due to the fact that OBS already down samples the audio, correct?

Thanks!

That finally showed up! I got the notification e-mail immediately, but it wasn't on the site until now. I just figured you'd deleted it.

Anyway, yes, it's better to keep the same clock all the way through the system. Note that that's the same *clock*, and not just the same number. See my note about that in my first reply. Pro rigs that have a bunch of things to keep in sync, have a way to distribute a master clock to everything else. Originally, that was a dedicated coax that only did that, and some of the old gear that I've accumulated has the BNC jack for it; not sure what's done now.

If you can't guarantee the same *clock*, not just the same setting, then I wonder about the wisdom of forcing a resample by using different settings on purpose. The result won't be any better than the worst thing in the chain, but it might be better than a naive assumption that "clocks don't drift", and the effects of that being wrong? I really don't know.

If the resampler also has that assumption, then you haven't solved anything, and added the problems of a bad resampler. Resampling done right is completely transparent, but not everyone gets it right, even the big players.