Broadcast audio and live audio are NOT the same thing! And of course, when you broadcast live, you have to translate in real-time, from one to the other. You can't just plug the two together and expect both the volume and the quality that you're used to from a commercial broadcast.
For live audio, you don't know what's coming, so you have to keep some "headroom", as we call it - turn it down a bit, right up front at the mic itself, or the physical thing the directly receives the raw mic - so that the natural spikes and variation don't overload it. If you mess up here, there's no fixing it later...but it also sounds quiet if you just push it out to your audience as-is.
For broadcast audio, you need to completely fill the meter. Historically, this was done to overcome the limitations of the distribution medium - vinyl, cassette tape, radio, etc. - and it's become the standard that everyone's used to and what their listening controls are set for. Again, if you did that with a raw mic, you'd absolutely kill the quality. Practically clipping *all the time*.
In between, to do that translation, you need a compressor. Often followed by a limiter. OBS has both as filters. Both are like fast automatic volume controls, so you can keep the average high, but it turns down the peaks and comes right back up again for the rest. No way you could do all that by hand! OBS has terrible visibility for its tools, so here's a screenshot of a much better one:
The graph on the right is what OBS is sorely missing! It's a plot of the input volume horizontally vs. the output volume vertically. As you can see, it's a normal boost at low volume, but then it flattens out at high volume. The transition is called the Knee.
In order of OBS's settings:
- Ratio controls how flat the high-volume part of the curve is.
- A higher ratio squashes more (flatter line), and produces a more constant output volume regardless of the input volume, but only if the input is above the Threshold.
- Threshold controls the input volume where the compressor starts working at all.
- Below this, it's just a normal boost as set by the Output Gain.
- Above this, it progressively turns itself down as set by the Ratio.
- Attack is how fast it turns itself down for a louder input.
- Release is how fast it turns itself up for a softer input.
- Output Gain (called Makeup in the screenshotted one) is the amount of boost that it has to start with. The compressor action itself turns down from here.
- If you're keeping a "live level" at this point in the processing chain, you'd watch the meter that shows how much it's turning down, and set the Gain to match that...except that OBS doesn't have any of those meters! Which is why I'm using the screenshot of a different one.
- If you're translating at this point from "live" to "broadcast", you'd use the final output meter to set the Gain to *exactly* fill it. This is annoyingly the *only* meter that OBS has.
- Sidechain selects which audio signal controls the compressor, if it's different from the one that it's processing.
- "None" uses the same signal (called Feed-forward in the screenshot, because that one also has a Feed-back option and a few others)
- Anything else allows *that* signal to drop the volume of *this* one according to this one's settings, without affecting that other one, and without actually compressing this one. This might be used to have some background music at full volume, for example, without interfering with your voice, all automatically. In that case, you'd put the compressor on the music, and set its sidechain to your voice.
- If you want to both compress and "duck", as it's called, you'd need *two* compressors:
- The first in the chain set to "None", as the compressor.
- The second set to the other signal, as the ducker.
Other significant controls shown here, that OBS does not have:
- Knee is how gradually the two lines connect:
- Hard knee (0dB radius) is an instant transition. Very audible if you "ride the knee", not so much for incidental spikes.
- Soft knee (non-zero radius) is a more gradual transition.
- A *very* soft knee as shown here, combined with a high ratio, is somewhat like an "automatic ratio". At low volume, it might do "something", but you'd be hard-pressed to notice. As the input volume increases, it slowly clamps down harder and harder until it becomes practically a limiter. I like that because it keeps most of the transparency for normal speaking, and gradually transitions into "definitely being controlled" for yelling. But OBS doesn't have that soft of a knee, or even a control for it at all.
- To approximate a soft knee, you might have several compressors, one after another, with progressively higher ratios...
- The Mix settings allow for "parallel compression" by mixing the original input signal with what the compressor comes up with.
- This effectively boosts softer signals via the Wet or compressed path, without squashing the louder sounds because the original Dry path takes over then. OBS doesn't do that...unless you have multiple copies of the same source, one with the compressor and one without.
Anyway, to send a live mic to broadcast, all in OBS, you might want something like this:
- Noise Suppressor, as the FIRST thing in the chain, so it's not trying to chase a varying noise floor after the compressors.
- Other processing as desired.
- Compressor
- Ratio between 2:1 and 6:1
- Threshold to make it work *most* of the time but not all the time
- Attack and Release set by ear to sound natural or transparent
- Output Gain to get back to the original "live" volume with normal material
- Compressor
- Ratio 10:1
- Threshold to make it work about half as much as the first one
- Attack and Release set by ear to sound natural or transparent
- Output Gain to exactly fill the meter
- Limiter
- Threshold at -1dB
- Release set by ear to sound natural or transparent
#2 and #3 might be swapped or interleaved as desired, but the last two need to be in that order and LAST. Likewise, the first one does need to be FIRST, or at the very least, before any compressors, limiters, gates, or other dynamic things.
Record yourself dead raw, then play that through your speaking chain. Set each thing in order, with everything after it either not added yet or disabled. Turn your speakers up as needed, until you get to the point where the processing chain does that for you.
---
For one of my rigs, that has the audio path entirely *outside* of OBS and only gives it the finished soundtrack to pass through unchanged, I have these two compressors, preceeded by a Noise Suppressor and a 2nd order highpass at 150Hz:
Highpass -> Noise Suppressor (order doesn't really matter for those two specifically) -> Left Compressor here -> Right Compressor here.
I'm using the right compressor as my Limiter, because that's all a limiter is: just a compressor with the Ratio all the way up, and usually a hard Knee and instant Attack, which I've also set here. If you take the vertical axis of the left one and put it on the horizontal axis of the right one, you might see that a typical raw input level of -18dBFS or so, "rides the knee" of the left (first) one, which puts it right at the knee of the second but doesn't quite get there, which puts the final output right at full-scale.
Whispering (farther left on the left graph) doesn't reduce the final output by much, but it does a little bit. Much more importantly though, it's not compressed as much (lower ratio because of the really soft knee that it's riding), so it sounds like a bigger difference than it really is.
Yelling will never go over full-scale either. That'll ride up higher on the super soft knee, which is a high ratio and thus more compressed. Again, the aggressive compression is what makes it sound louder, not the actual transmitted volume that barely changed at all.
You might also note that the Attack time is not quite instant for the first one that does most of the work. 10 milliseconds, or 1/100th of a second, which corresponds to 1/2-wave at 50Hz. Low frequencies will be distorted, starting slightly higher than that. So pretty much all of what the highpass allows through, stays clean at this stage, and the more "peaky" parts of words stay "peaky" too, which makes it sound a little bit more natural.
The second one, as a Limiter, *is* instant. Its output will *never* exceed the set level, even if it has to distort to do it. But because it's just a momentary thing - clipping only the single first wave, and all the rest stay clean at a low enough volume that fits in the meter - you're not going to notice.
