Best OBS Noise Suppression Settings (August 2026) RNNoise vs Speex

If your stream sounds like a jet engine took off next to your microphone, your OBS noise suppression settings are the first thing to fix. I have helped dozens of streamers clean up their audio over the past few years, and the single biggest win comes from understanding the difference between RNNoise and Speex. Both are built into OBS Studio, both are free, and both genuinely remove background noise. But they do it in very different ways, and choosing the wrong one (or the wrong settings) can leave your voice sounding muffled, robotic, or cutting out entirely.

This guide breaks down exactly how RNNoise and Speex work, which one you should pick for your use case, and the exact settings I recommend for streaming, podcasting, and gaming. I also include a complete OBS audio filter chain you can copy, plus troubleshooting for the most common audio problems people hit. By the end, you will know precisely which noise suppression method fits your setup and how to configure it from scratch.

What Is OBS Noise Suppression and Why It Matters?

OBS noise suppression is a built-in audio filter that removes unwanted background sound from your microphone signal before it reaches your stream or recording. Things like PC fan hum, keyboard clicks, room echo, air conditioning, and street noise all count as background noise. Even the best microphones pick up some of this, and viewers notice immediately when audio is messy. Research consistently shows that audiences will tolerate mediocre video but will mute a stream within seconds of bad audio.

OBS Studio ships with two different noise suppression methods you can switch between: RNNoise and Speex. RNNoise is a machine learning model trained on thousands of voice samples. It analyzes your audio in real time and separates human speech from everything else using a recurrent neural network. Speex is an older, lighter algorithm that uses spectral subtraction to estimate and remove noise. Both work in real time and require no internet connection, but their quality and CPU usage differ significantly.

You only need noise suppression if your recording environment has audible background noise. If you record in a professionally treated studio, you can skip it entirely. For most home streamers and creators, though, some form of noise suppression is essential. The question is which method to use, and that depends on your microphone, your CPU, and how aggressive you want the filter to be.

RNNoise vs Speex: Core Comparison

RNNoise generally produces better sound quality with less voice distortion, while Speex uses noticeably less CPU. That is the headline difference, and it is the one OBS’s own documentation confirms. But the full picture involves a few more tradeoffs you should understand before picking one for your stream.

RNNoise runs through a small neural network trained on real human speech and noise samples. This is why it sounds more natural. It can tell the difference between your voice and a similar-sounding noise, such as a fan at the same pitch. Speex, by contrast, just removes any frequency that looks like noise, which means it sometimes chops out parts of your voice too.

In my testing, RNNoise uses roughly 2-5% of a modern quad-core CPU during normal speech. On a six-core or eight-core processor, this is essentially free. Speex sits well under 1%. If you are streaming on a budget laptop or an older CPU and you notice frame drops when filters are enabled, Speex is the safer pick. Otherwise, RNNoise wins on quality almost every time.

How to Set Up RNNoise and Speex in OBS (Step by Step)

Setting up either filter takes about two minutes. The interface is identical for both, so you can swap between them anytime to compare.

Setting Up RNNoise in OBS Studio

Open OBS Studio and click the Settings button in the bottom-right corner of the Audio Mixer panel. Select the Audio tab and make sure your Sample Rate is set to 48 kHz. RNNoise needs this sample rate to work properly.

Next, click the three-dot menu next to your microphone in the Audio Mixer and select Filters. Click the plus icon in the bottom-left and choose Noise Suppression from the dropdown. In the filter dropdown, select RNNoise. Leave the suppression level at the default and click OK.

Speak into your microphone and check the audio meter. You should see the same activity as before, but the background noise floor should now sound noticeably quieter when you stop talking. If your voice sounds muffled, check the troubleshooting section below.

Setting Up Speex in OBS Studio

Follow the same steps as above, but choose Speex instead of RNNoise in the Method dropdown. Speex gives you a Suppression Level dropdown with three options: Low, Medium, and High. Start with Medium and only move to High if your background noise is extreme.

High Speex suppression is where most of the robotic-sounding audio comes from. The algorithm is too aggressive at that level and starts cutting into your voice. If you find yourself needing High, you are usually better off with RNNoise instead, or fixing the noise at the source.

The Optimal OBS Audio Filter Chain

Noise suppression alone rarely produces broadcast-ready audio. The community-tested filter chain that produces the cleanest results in OBS is: Noise Suppression, then Compressor, then Gain, then Limiter. Each filter handles a specific job, and the order matters.

1. Noise Suppression (RNNoise) – Removes constant background noise like fans and hum.

2. Noise Gate – Mutes your mic completely when you stop talking, eliminating residual room noise.

3. Compressor – Evens out your volume so quiet speech and loud speech sit closer together.

4. Gain – Boosts your overall level back up after compression reduces it.

5. Limiter – Caps the loudest peaks to prevent clipping and harsh distortion.

Here are the specific settings I recommend for each filter in this chain. They are tuned for a typical USB condenser microphone in a moderately quiet room. Adjust them based on your mic and environment.

The noise gate is the filter many people skip but absolutely should not. Even great noise suppression leaves a faint hiss or room tone when you stop talking. A noise gate with a Close Threshold of -32 dB and an Open Threshold of -26 dB creates a 6 dB gap that prevents the gate from opening and closing constantly on quiet sounds. The Hold Time of 200 ms keeps the gate from snapping shut between syllables.

The compressor ratio of 3:1 is a starting point. If you speak very dynamically (loud laughter, quiet asides), bump it to 4:1. If your voice already sounds consistent, 2:1 is plenty. The goal is to keep your audio around -10 to -6 dB on the meter during normal speech without clipping when you raise your voice.

Recommended Settings by Use Case

Different use cases demand different noise suppression approaches. A podcast recording needs more aggressive cleanup than a casual gaming stream, and a music-sensitive broadcast needs different settings than a high-energy gameplay broadcast. Here are the specific configurations I recommend for each scenario.

Best Settings for Live Streaming

For live streaming on Twitch or YouTube, viewer retention depends on clean, intelligible audio. Use RNNoise with the standard filter chain above. Set your compressor threshold to -18 dB with a 3:1 ratio. Your limiter should sit at -1 dB to prevent sudden shouting from clipping. This gives viewers a consistent, easy-to-listen-to broadcast without sounding overprocessed.

Best Settings for Podcasting

Podcast audio needs to be broadcast-quality because listeners have higher expectations and there is no visual to distract from audio issues. Use RNNoise plus a slightly more aggressive noise gate (Close Threshold -36 dB). Add an Expander filter after the Noise Gate with a ratio of 2:1 and a threshold of -50 dB. The expander pushes background noise even further down between phrases without the abrupt cut of a gate.

Best Settings for Gaming Streams

Gaming streams often involve keyboard and mouse clicks, which are the hardest sounds to remove without affecting your voice. RNNoise handles these better than Speex because it can distinguish transients like keystrokes from sustained tones. Use the standard filter chain but lower the noise gate Close Threshold to -28 dB so rapid in-game reactions are not cut off. Keep messages to your team clearly audible.

For all three use cases, the deciding factor is how loud your environment is. If you record in a genuinely quiet room, Speex on Medium is enough. If you have PC fans, AC, or street noise, RNNoise will save you. If you have all of those plus a mechanical keyboard, you need RNNoise plus a noise gate plus good mic placement.

Microphone Type Matters: Dynamic vs Condenser

The type of microphone you use changes how your noise suppression behaves. Condenser microphones (like the Blue Yeti or AT2020) are more sensitive and pick up more room noise, which means they benefit more from RNNoise. Dynamic microphones (like the Shure SM7B or Samson Q2U) reject more background noise naturally, so they can often get away with Speex or even no noise suppression at all.

If you use a condenser mic, position it as close to your mouth as possible (4-6 inches) and slightly off-axis. This reduces how much room noise the capsule picks up and lets the noise suppression filter work less aggressively. If you use a dynamic mic, you can usually keep your noise suppression on the lighter setting and focus more on the compressor and limiter for broadcast quality.

Lavalier and headset microphones fall somewhere in between. They sit close to your mouth but their small capsules often have more self-noise. RNNoise handles this self-noise better than Speex, which tends to treat the hiss as part of the voice signal.

NVIDIA Noise Removal as a Third Option

If you have an NVIDIA RTX graphics card, OBS Studio now includes NVIDIA Noise Removal as a third option. It runs on your GPU instead of your CPU, which frees up processing power for encoding. In testing, it produces cleaner audio than RNNoise but at the cost of slightly more GPU usage. It is worth trying if you stream on a mid-range CPU and have headroom on your GPU.

You cannot run RNNoise and NVIDIA Noise Removal at the same time. Pick one. If you have both options available, I recommend NVIDIA for heavy noise environments and RNNoise for moderate noise. The setup is identical: choose Noise Suppression from the Filters menu and select NVIDIA from the dropdown.

Troubleshooting Common Audio Issues

Even with the right settings, you can still hit audio problems. Here are the most common issues and how to fix them.

Why Your Voice Sounds Muffled After Noise Suppression

This is the most common complaint about RNNoise. The filter is removing high-frequency content from your voice along with background noise. To fix it, move the microphone closer to your mouth (4-6 inches), add a small boost to your Gain filter (+3 to +6 dB), and consider adding an EQ filter after the noise suppression to restore the missing high frequencies. A simple high-shelf boost at 8 kHz with +2 dB usually does the trick.

Why RNNoise Cuts Out Your Voice When You Speak Quietly

RNNoise occasionally misclassifies quiet speech as noise and suppresses it. This happens more often with higher-pitched voices and softer speakers. The fix is to increase your microphone gain in Windows or your audio interface so RNNoise sees a stronger signal. If you are already at max gain, the next step is to enable the Compressor first in your filter chain so your quiet voice is boosted before noise suppression analyzes it.

Why Keyboard Noise Still Comes Through

Mechanical keyboards produce sharp transient sounds that look similar to speech sibilants. RNNoise catches most of these, but very loud keyboards can leak through. The fix is a noise gate with a faster attack time (3 ms) and a slightly higher Close Threshold (-28 dB). You can also enable keyboard acoustic dampening or switch to a quieter switch type like tactile or linear instead of clicky.

Why Muting Your Mic Spikes CPU Usage With RNNoise

This is a known OBS bug. RNNoise continues processing audio even when your mic is muted, and render times spike as a result. The workaround is to either switch to Speex (which does not have this issue) or unmute your mic and rely on a noise gate to handle silence. Community tests confirm this spike happens specifically with RNNoise plus mute, not with Speex plus mute.

Why Noise Suppression Sounds Robotic

Robotic audio happens when the filter is over-suppressing. If you use Speex, switch from High to Medium. If you use RNNoise, it is usually about input gain. Lower your microphone gain to -10 dB or so and raise the post-filter Gain to compensate. Over-driven inputs cause both RNNoise and Speex to misbehave.

Frequently Asked Questions

Final Verdict: Which Should You Choose?

For most streamers and content creators, RNNoise is the better choice. It produces cleaner, more natural voice audio with less robotic distortion, and it handles tricky sounds like keyboard clicks and room echo better than Speex. The CPU cost is small on modern hardware, and unless you are on a budget laptop or running on integrated graphics, you will not notice it.

Choose RNNoise if you have a modern CPU, want the best voice quality, and your main concern is audience experience. Choose Speex if you are on a low-end CPU, have a quiet environment and just need light cleanup, or if RNNoise is causing render time spikes when you mute your mic. Choose NVIDIA Noise Removal if you have an RTX card and your CPU is the bottleneck in your stream.

Whichever method you pick, pair it with the full filter chain from this guide, set your sample rate to 48 kHz, and spend time on microphone placement. The best noise suppression in the world cannot fix a microphone that is six inches too far from your mouth. With the right setup, you will have broadcast-quality audio that keeps viewers watching and listeners subscribed.