OBS Audio Filter Chain Settings for a Clean Stream Mic
Someone sent me a 40 minute VOD last month and asked why chat kept telling her the mic was both too quiet and too loud. She was right to be confused.

Someone sent me a 40 minute VOD last month and asked why chat kept telling her the mic was both too quiet and too loud. She was right to be confused. Her voice sat around 32 dB below full scale while she talked normally, then jumped close to clipping every time she laughed, and the room air conditioner hummed underneath the whole thing. She had five filters stacked in OBS. Four of them were fighting each other.
The filter chain is the part of a stream setup people copy from a screenshot and never revisit. That is how you end up with a noise suppressor set to maximum, a gate that cuts the first syllable off every sentence, and a compressor doing almost nothing because the signal reaching it is already squashed. The settings matter, but the order matters more, and the order is not a matter of taste.
Order first, numbers second
OBS runs filters top to bottom in the list, so each one hears whatever the one above it produced. Put a compressor above a gate and the compressor lifts your room noise toward the gate threshold, which makes the gate chatter open and shut on nothing. Put noise suppression at the very top and it works on a signal that is still too quiet for it to tell voice from hiss.
The order I use on every stream mic, from top to bottom: gain, noise suppression, gate, compressor, limiter. Gain brings the signal into a sensible window so everything below it has something to work with. Suppression shaves the steady hum. The gate closes the channel in the gaps. The compressor evens out the range of what is left. The limiter catches the one unexpected shout.
If you only change one thing today, change the order. I have fixed three streams this year by dragging the gate above the compressor and touching nothing else.
The numbers I start from
These are starting points, not answers. A Rode PodMic at 99 dollars into a cheap interface needs far more gain than an Elgato Wave:3 at 159 dollars, and a treated corner needs a gentler gate than a bedroom with a laptop fan in it.
| Filter | Setting | Starting value |
|---|---|---|
| Gain | Boost | Whatever puts normal speech near 18 dB below full scale |
| Noise suppression | Method | RNNoise, or Speex at 30 dB on a weak CPU |
| Noise gate | Close threshold | 44 dB below full scale |
| Noise gate | Open threshold | 36 dB below full scale |
| Noise gate | Attack time | 5 ms |
| Noise gate | Hold time | 200 ms |
| Noise gate | Release time | 150 ms |
| Compressor | Ratio | 4:1 |
| Compressor | Threshold | 22 dB below full scale |
| Compressor | Attack | 8 ms |
| Compressor | Release | 120 ms |
| Limiter | Threshold | 2 dB below full scale |
| Limiter | Release | 60 ms |
Two of those values do most of the work. The 8 ms gap between the gate open and close thresholds is what stops the chatter: a gate with a single threshold flutters on any sound that hovers right at the line, and a stream of flutter is more distracting than the hum you were trying to remove. The 200 ms hold is what stops the gate slamming shut between words in a slow sentence.
The compressor threshold is the one people set wrongly most often. If your gain reduction meter is pinned at 12 dB the whole time you speak, the threshold is too low and you have turned a dynamics tool into a volume control. Look for 3 to 6 dB of reduction on normal speech and 10 or so on a laugh.
Noise suppression is oversold
Every stream guide I read recommends switching noise suppression on and sliding it as far as it goes. I would push back on that, and I say so as someone who learned this on community radio desks where there was no suppression plugin to hide behind.
RNNoise is a model trained on speech. When you push it hard it removes the parts of your voice that resemble noise, which means breath, sibilance and the tail of every word. The result is intelligible and slightly dead, like a phone call on a good day. On a 90 minute stream that dullness accumulates, and viewers do not report it as a problem. They just leave earlier.
A moving blanket on the wall behind your monitor and a mic 15 cm from your mouth will beat any suppression setting. Thirty dollars of sound absorption at the first reflection point did more for one client's VOD than three weeks of plugin fiddling. Where suppression earns its place is a steady mechanical hum you genuinely cannot move away from, a fridge compressor two metres away or a desktop under load. Switch it on, use the lightest setting that kills the hum, and leave the rest to the room. Our piece on taming a boomy recording room applies to a stream desk almost unchanged.
Setting the gate in four minutes
Open the audio mixer, switch the meters to show peak and hold, and sit silent for 20 seconds with everything in the room running as it does during a stream. Note where the noise floor settles. Mine, in a carpeted spare room with a PC two metres away, sits around 52 dB below full scale.
Set the gate close threshold 8 dB above that floor and the open threshold 8 dB above the close. Then read a paragraph out loud at your real stream volume, which is louder than your test voice, and watch for two failures. If the first consonant of a sentence disappears, lower the open threshold by 3 dB. If you can hear the room appear and vanish behind your voice, you are hearing the gate work, and the fix is a longer release rather than a deeper cut.
Test it with a keyboard too. Mechanical switches land in the same range as speech transients, so a gate tuned only on your voice will open for every keypress during a quiet moment. If that happens, the honest answer is a different mic position rather than a harsher gate.
What to watch once you are live
I keep two things on screen. The gain reduction meter on the compressor, because a meter that never moves means the threshold is wrong and a meter that never rests means the same thing in the other direction. And the peak meter, because a stream that lives above 6 dB below full scale has no headroom left for a sneeze.
Platform loudness normalisation deserves a mention here, since it undoes a lot of what streamers do to sound loud. Live audio is not normalised the way an uploaded video is, but the VOD that gets published afterwards often is, and a mix crushed to a very high average simply gets turned down with its dynamics already gone. Aim your spoken level so the overall mix lands near 16 LUFS integrated rather than chasing the loudest possible signal, and check the number on the recording afterwards. The same logic I lay out in loudness targets for video holds for anything you archive.
One last habit worth building: record a local copy and listen to the first five minutes on a phone speaker. Filter problems that are inaudible on good headphones are obvious on a small driver, and the gate chatter you missed will be the first thing you hear. More of this sits in the stream setup section.
FAQ
Do I need all five filters?
No. A quiet room with a decent dynamic mic often needs only gain, compressor and limiter. Add the gate when the room has a floor you can hear, and the suppressor only when something mechanical refuses to go away. Fewer filters usually sounds better than a long chain set carelessly.
Why does my voice sound thin after adding the compressor?
Usually the threshold is far too low, so quiet detail and loud peaks are squeezed into the same narrow band. Raise the threshold until the gain reduction meter rests at zero between sentences. A high pass filter cutting below 80 Hz can also thin a voice if it is set at 150 Hz instead.
Should the same chain apply to my guests on a call?
Not the same values. A remote guest arrives already processed by their own platform, so a second compressor stacks on top of theirs. Give the guest track its own light chain, usually gain and a limiter, and nothing else.
Does a hardware processor beat the OBS filters?
For most people, no. A dynamics processor built into an interface frees a little CPU and gives you a knob to grab mid stream, which matters if you switch between talking and shouting. The sound difference is small enough that I would spend the money on acoustic treatment first.
More in Stream Setup
Stream Setup
Recording a Local Backup in OBS While You Stream Live
A friend streamed a two hour board game session, promised the group a highlight reel, then opened the platform VOD the next...
Stream Setup
Two PC Streaming Setup and When It Is Worth Building
Your game stutters every time the stream encoder spikes, or your viewers see blocky motion whenever the action gets busy.
Stream Setup
Green Screen Setup for Streaming in a Small Room
Your chroma key looks fine until you move, and then your hair flickers, your shoulders grow a green halo and a chunk of your...