Audio Issues related to Peaking

[Archedas]

Recently, been having an issues with losing audio to stream whenever particularly loud/high pitched sounds go through the mixer. While it does not show as peaking in the Mixer itself, obs appears to be losing audio output and going entirely silent. I know for sure it's not related to anything CPU usage wise, as we rarely are hitting above 15-20% used by OBS.

We are using a Behringer X2222USB Mixer, fed into the PC via a single USB cable, with 6 XLR connected Audio-Technica ATR2100-USB Microphones.

Currently, the only 'solution' to this I've been able to test on my own without having 6 people speaking/laughing at once is to tank the Gain on each mic Mixer to about 25% with Compression at about 50% and the high frequency EQ at around 55-65% (unfortunately I can't give exact numbers, as the mixer doesn't have a digital feed on what the knobs are set to exactly, just generally position along their range). I then have to add about 17.00db of Gain via a Filter in OBS.

We have the VST plugin options with a Noise-Removal profile setup to remove a light hiss in the mic, but on or off the clipping issue still exists so I don't believe that's part of the problem.

With the above changes (Gain ~25%, Comp ~50%, High EQ ~55-65% on Mixer, +17.00db Gain in OBS filter, VST Noise Removal Plugin) I can cause the obs audio to show as over 0.0db / peaking while yelling into the mic and using high pitched voices, but the audio no longer appears to be cutting out when such happens, both in a local recording or stream (tested both ways to ensure no encoding or decoding issues were the culprit).

I did my testing with a singular mic run solo - the issues we have can be triggered solo, but are most often occurring with 6 individuals laughing or speaking into the mic around the same time. Large bursts of audio or high pitched voices seem to trigger things more often.

I have considered a Limiter set at -1.50db but I still can notice it starting to cut or having odd distortion when I get around that level.

Other recommendations on things to attempt to handle sudden bursts of higher then normal volume without causing major distortion or audio to be cut would be much appreciated.

 

[AaronD]

Recently, been having an issues with losing audio to stream whenever particularly loud/high pitched sounds go through the mixer. While it does not show as peaking in the Mixer itself, obs appears to be losing audio output and going entirely silent.

Does OBS lose the input too?

We are using a Behringer X2222USB Mixer, fed into the PC via a single USB cable, with 6 XLR connected Audio-Technica ATR2100-USB Microphones.

Currently, the only 'solution' to this I've been able to test on my own without having 6 people speaking/laughing at once is to tank the Gain on each mic Mixer to about 25% with Compression at about 50% and the high frequency EQ at around 55-65% (unfortunately I can't give exact numbers, as the mixer doesn't have a digital feed on what the knobs are set to exactly, just generally position along their range). I then have to add about 17.00db of Gain via a Filter in OBS.

Where are those percentages measured from? Always the bottom/left end of travel? Referenced to nominal for each? (EQ nominal is in the middle)

A picture of the board might be helpful to avoid that ambiguity. And you're right, the USB connection doesn't know the settings of anything; it's just a jellybean stereo sound card shoved into an otherwise pure-analog design. (the digital FX unit is the same way)

Another thing you could do, to create a loud signal all by itself, is patch an unused output back to an input, then send that input to that output. That's an intentional feedback loop, and if you get the total loop gain above unity, it'll oscillate on its own, full-scale. Then you can send it somewhere else too, and even shape the sound with the preamp and EQ to use as an effect if you want.
You're not going to break it, since the output is running from the same power supply as the input, and therefore can't overdrive it. The preamp *will* clip, and possibly the EQ too, but only because it's trying to produce a signal that is larger than that supply. At no point is anything in danger...unless you involve some outboard gear in that loop.

I have considered a Limiter set at -1.50db but I still can notice it starting to cut or having odd distortion when I get around that level.

Other recommendations on things to attempt to handle sudden bursts of higher then normal volume without causing major distortion or audio to be cut would be much appreciated.

If you're going to use a limiter to avoid clipping, then the *input* to that limiter must already be low enough to never clip, and you use the limiter's gain to bring it up to your desired level.

 

[Archedas]

Does OBS lose the input too?


Where are those percentages measured from? Always the bottom/left end of travel? Referenced to nominal for each? (EQ nominal is in the middle)

A picture of the board might be helpful to avoid that ambiguity. And you're right, the USB connection doesn't know the settings of anything; it's just a jellybean stereo sound card shoved into an otherwise pure-analog design. (the digital FX unit is the same way)

Another thing you could do, to create a loud signal all by itself, is patch an unused output back to an input, then send that input to that output. That's an intentional feedback loop, and if you get the total loop gain above unity, it'll oscillate on its own, full-scale. Then you can send it somewhere else too, and even shape the sound with the preamp and EQ to use as an effect if you want.
You're not going to break it, since the output is running from the same power supply as the input, and therefore can't overdrive it. The preamp *will* clip, and possibly the EQ too, but only because it's trying to produce a signal that is larger than that supply. At no point is anything in danger...unless you involve some outboard gear in that loop.


If you're going to use a limiter to avoid clipping, then the *input* to that limiter must already be low enough to never clip, and you use the limiter's gain to bring it up to your desired level.

OBS does not lose the input - it just continually shows the microphone input levels as a solid red bar when it occurs.

Gain measured from 0, and I suppose with EQ being measured from the middle of its levels, Compression at nominal, and High EQ about 10% higher then others. Mid and Low remain nominal.

I've placed red marks in the approximate locations of where the knobs are turned currently in the attached image in the first channel of the mixer.

 

[AaronD]

Compression at nominal...

I've placed red marks in the approximate locations of where the knobs are turned currently in the attached image in the first channel of the mixer.

If your goal is loud (for testing), then I think the compressor on the board is actually hurting you. Yes, it turns up quiet sounds, which might have given you the idea, but it also *turns down* loud sounds. The higher you turn that knob, the harder it is to get above the nominal operating level of that board.

...losing audio to stream whenever particularly loud/high pitched sounds...obs appears to be losing audio output and going entirely silent.

Do you have a Compressor or Limiter filter in OBS? Your description sounds to me like you have an aggressive one with a fast attack and really slow release. If so, then it won't be *entirely* silent, but compared to what just came before it, it can seem that way, before fading back in.

If you use my self-oscillator trick, then you can play with slowly increasing the volume to see if it's an abrupt change all by itself, or if you have to give it an abrupt change first.
If you're not using Aux 1 and Mic 8, for example, you can run a short cord between those two jacks. Your feedback loop then includes the Preamp, Compressor, *maybe* EQ (*), Aux 1 Send, and Aux 1 Master. If you're already using Aux 1 but not Aux 2, same thing, but make sure that it's set to Pre-Fade, next to the Send knob. The loop then operates on its own, freerunning, and you can use the fader and routing buttons to make a slow or abrupt transition without requiring the loop itself to restart and stabilize.

(*) The manual is missing the most important page: the overall diagram.

 

[Archedas]

If your goal is loud (for testing), then I think the compressor on the board is actually hurting you. Yes, it turns up quiet sounds, which might have given you the idea, but it also *turns down* loud sounds. The higher you turn that knob, the harder it is to get above the nominal operating level of that board.


Do you have a Compressor or Limiter filter in OBS? Sounds to me like you have an aggressive one with a fast attack and really slow release. If so, then it won't be *entirely* silent, but compared to what just came before it, it can seem that way.

If you use my self-oscillator trick, then you can play with slowly increasing the volume to see if it's an abrupt change all by itself, or if you have to give it an abrupt change first.
If you're not using Aux 1 and Mic 8, for example, you can run a short cord between those two jacks. Your feedback loop then includes the Preamp, Compressor, *maybe* EQ (*), Aux 1 Send, and Aux 1 Master. If you're already using Aux 1 but not Aux 2, same thing, but make sure that it's set to Pre-Fade, next to the Send knob. The loop then operates on its own, freerunning, and you can use the fader and routing buttons to make a slow or abrupt transition without requiring the loop itself to restart and stabilize.

(*) The manual is missing the most important page: the overall diagram.

We currently have nothing in OBS filters except a noise removal and gain filter added to accommodate the low gain on the Mixer and get audio levels where we went them for a normal speaking volume.

Mostly, we're trying to get settings to allow for a normal speaking volume that can handle sudden bursts of louder sounds or multiple inputs at once (such as all 6 mics having suddenly louder sounds) all feeding in when normally it's balanced around 1-2 maximum at a time.

This is mostly to handle unexpected bursts in volume and higher voices / everyone laughing at once - An example of the issue we are having can be heard around the 56:55 minute mark for a few minutes on and off, and again around 1:34:00 in this twitch vod: https://www.twitch.tv/videos/1831776333

 

[Archedas]

We currently have nothing in OBS filters except a noise removal and gain filter added to accommodate the low gain on the Mixer and get audio levels where we went them for a normal speaking volume.

Mostly, we're trying to get settings to allow for a normal speaking volume that can handle sudden bursts of louder sounds or multiple inputs at once (such as all 6 mics having suddenly louder sounds) all feeding in when normally it's balanced around 1-2 maximum at a time.

This is mostly to handle unexpected bursts in volume and higher voices / everyone laughing at once - An example of the issue we are having can be heard around the 56:55 minute mark for a few minutes on and off, and again around 1:34:00 in this twitch vod: https://www.twitch.tv/videos/1831776333

Note: when these occured, we had different settings. The settings I have above in the image are my attempts to fix the issue so far.

During that recording we had no gain added, but had the RNNoise Noise Supression filter enabled, which I think was part of the issue.

 

[AaronD]

We currently have nothing in OBS filters except a noise removal and gain filter added to accommodate the low gain on the Mixer and get audio levels where we went them for a normal speaking volume.

Mostly, we're trying to get settings to allow for a normal speaking volume that can handle sudden bursts of louder sounds or multiple inputs at once (such as all 6 mics having suddenly louder sounds) all feeding in when normally it's balanced around 1-2 maximum at a time.

This is mostly to handle unexpected bursts in volume and higher voices / everyone laughing at once - An example of the issue we are having can be heard around the 56:55 minute mark for a few minutes on and off, and again around 1:34:00 in this twitch vod: https://www.twitch.tv/videos/1831776333

Hmm...well, I guess the Noise Suppressor could be getting confused with the laughter and the weird voice. It's really designed for a business meeting or something like that, in a conference room with loud air conditioning, or in a lobby with background music that needs to be removed.

How much noise do you actually have, if you take off the Suppressor?

And, how much noise do you gain if you add a Compressor and Limiter filter in OBS? (in that order) Your audio is still pretty quiet, despite hitting the top of the meter, and that's the difference. Squash the individual mics (turn their compression knobs way up on the board), then squash the master too. Since your board doesn't have a master compressor, you'll have to do that in OBS. (unless maybe your FX unit has one?)

So:

• Disable the Noise Suppressor. Don't remove it just yet, but you might eventually.
• Add a Compressor. Set it to handle *most* of the variation but not all of it, and use its makeup gain to get back about as much volume as it takes off on average. You want it to do its job without being noticed.
• Add a Limiter after the Compressor. Set this one as a "hard-stop safety-net" that you're going to hit every once in a while, but not all the time. The level coming out of this one should be right at full-scale peak without ever going over, and about -10dB to -15dB average (the black bar across the meter).

That should get you enough level that your listeners don't have to turn their volumes up for you...and then forget to turn back down for the radio! But it also amplifies noise. So how much noise can you afford? I thought your sample sounded pretty good; based on that, you can afford to have at least a *little* bit more noise.

Also, what happens if you pan the mics away from center? So the one on the far left comes mostly from the left speaker, the one on the far right comes mostly from the right speaker, and those in between are in similar positions in the "soundstage"? I wouldn't pan ALL the way to one side or the other, 'cause that's pretty disorienting in headphones, but maybe +/- 90deg or less from center on the pan knob? (leave the last 45deg unused)

 

[Archedas]

Hmm...well, I guess the Noise Suppressor could be getting confused with the laughter and the weird voice. It's really designed for a business meeting or something like that, in a conference room with loud air conditioning, or in a lobby with background music that needs to be removed.

How much noise do you actually have, if you take off the Suppressor?

And, how much noise do you gain if you add a Compressor and Limiter filter in OBS? (in that order) Your audio is still pretty quiet, despite hitting the top of the meter, and that's the difference. Squash the individual mics (turn their compression knobs way up on the board), then squash the master too. Since your board doesn't have a master compressor, you'll have to do that in OBS. (unless maybe your FX unit has one?)

So:

• Disable the Noise Suppressor. Don't remove it just yet, but you might eventually.
• Add a Compressor. Set it to handle *most* of the variation but not all of it, and use its makeup gain to get back about as much volume as it takes off on average. You want it to do its job without being noticed.
• Add a Limiter after the Compressor. Set this one as a "hard-stop safety-net" that you're going to hit every once in a while, but not all the time. The level coming out of this one should be right at full-scale peak without ever going over, and about -10dB to -15dB average (the black bar across the meter).

That should get you enough level that your listeners don't have to turn their volumes up for you...and then forget to turn back down for the radio! But it also amplifies noise. So how much noise can you afford? I thought your sample sounded pretty good; based on that, you can afford to have at least a *little* bit more noise.

Also, what happens if you pan the mics away from center? So the one on the far left comes mostly from the left speaker, the one on the far right comes mostly from the right speaker, and those in between are in similar positions in the "soundstage"? I wouldn't pan ALL the way to one side or the other, 'cause that's pretty disorienting in headphones, but maybe +/- 90deg or less from center on the pan knob? (leave the last 45deg unused)

The microphones there unfortunately have fairly short and very directional range - they drop off heavily in volume if moved more then 4-6 inches from, and don't pick up side-audio well - basically, we have to be speaking directly into the front of the mic or it won't pick up at a good input volume. We have to orbit the microphone to face each other in order to keep it directly in front of where we are talking from.

I'll give the others things a test tomorrow - Thanks for the suggestions so far.

 

[AaronD]

The microphones there unfortunately have fairly short and very directional range - they drop off heavily in volume if moved more then 4-6 inches from, and don't pick up side-audio well - basically, we have to be speaking directly into the front of the mic or it won't pick up at a good input volume. We have to orbit the microphone to face each other in order to keep it directly in front of where we are talking from.

What does that have to do with anything else? Except that the compressors have more to do? Did you misunderstand the panning, perhaps? That's a knob just above the fader, to bias it more towards one speaker or the other, in stereo.

I looked up those mics again. Cardioid. That's not *too* bad off-axis. It's when you start getting into super-cardioid and figure-8 that it drops off in a hurry towards the sides. If you cover the back ports like rappers like to do, you can make them omni, which may or may not help or hurt. (monitor engineers hate that, because it makes the floor wedges feed back when they didn't before) The reason that they have a non-omni pattern at all is because they rely on delayed cancellation, using the speed of sound around the mic itself as that delay.

For distance, you just need to boost the preamp is all...provided that you never clip it when you're close. I have a Behringer XM8500 in my studio, which is a cardioid designed for close-up (and loud) vocals, but I have it about 4 feet away at a normal speaking volume. Crank the preamp, and it's fine.

Every time you double the distance, you lose 6dB. So it's more sensitive to that when you're close, because it doesn't take much to double. That drop-off is simply because the same energy is spread over that much more area. (area of a sphere = 4*pi*(r^2), and it's the r^2 part that gives you 1/4-power, or -6dB, every doubling of r) For a line source, like a highway or train track, you only lose 3dB. (area of a cylinder, not counting the end circles = 2*pi*r*h, and the non-squared r gives you 1/2-power, or -3dB, every doubling of r) And for a plane source, like the wall of a room with something loud on the other side, you don't lose any with increasing distance. Of course, that only works if you still *have* a plane or a line at the distance that you're measuring. (reflectors, like other walls, floor, and ceiling, can increase the effective size in the same way that a mirror does for light) At some distance, every finite-sized structure in free space will effectively become a point, and consequently fall off at the 6dB rate.

There's also the proximity effect to consider, which applies for any non-omni pattern: closer = more bass, which is usually a problem. But sometimes the talent will use that as an intentional effect.
(it also applies to speakers that use the same principle to create directional patterns, like a cardioid subwoofer for example; close to one of those can get you thoroughly rearranged!)

 

[Archedas]

Turns out, 99% of our audio loss issue was due to the RNNoise Noise Removal Filter. We removed it, purged out OBS of all other filters, rebuilt the compressor we used with VST (very minimal settings meant only to handle sudden bursts of loud audio) and everything seems significantly better now.

We also setup a proper mixer-side and OBS-side audio monitoring and control for both desktop audio and studio-side volume controls, in addition to normalizing gain and comp on all mics.

Looks like everything is working so far - Just have to do a full test stream to see how things go.

 

[AaronD]

One of the pitfalls that even the pros fall into, is forgetting to compare what you're doing with either nothing at all, or what you had before. It's amazing how easy it is to spend a ridiculous amount of time fighting problems and getting it just right...only to hit the bypass button and have it sound 10x better immediately! (or to do all of that, and then see that it *was* in bypass the whole time!)