There is a particularly deflating moment after a good recording: you put on headphones and discover that the laptop fan has been presenting alongside you. The temptation is to drag every noise control as far as it will go. Please resist it. A voice that sounds watery, clipped, or abruptly switched on and off is often harder to listen to than a little honest room tone.
OBS can soften steady noise and keep quiet passages cleaner, but each tool solves a different problem. We will listen to the untreated microphone first, make one change at a time, and keep the version that still sounds like you. No universal preset can know your room, microphone, speaking distance, or voice.
Before touching a filter, listen to the noise
Steady fan, air-conditioning, or low room hiss: Noise Suppression is the natural first experiment. OBS describes it as a tool for mild background or white noise, not a rescue for a loud room.
Keyboard taps or breaths mostly between phrases: an Expander can turn quiet sounds down smoothly; a Noise Gate can close more decisively.
Noise remains underneath every spoken word: a gate cannot remove it while the gate is open. Improve the room or microphone position, then use modest suppression.
Buzz, crackle, clipping, or electrical hum: first inspect cables, ports, wireless interference, input gain, and the audio device. A software filter can disguise symptoms without repairing the signal path.
Other people, television, music, or a changing street: these overlap speech and are much harder to remove cleanly. Microphone placement and a quieter recording window usually beat more processing.
Win the easy acoustic fixes first
Noise falls with distance; your voice falls with distance too. Moving the microphone closer to your mouth improves the voice-to-room ratio before OBS has to guess what to preserve. Speak across the capsule rather than breathing directly into it, and use a pop filter when plosives become the next problem.
Move the microphone away from a laptop exhaust, desktop fan, open window, or vibrating desk surface.
Turn the microphone toward your mouth and, for directional microphones, point its rejection side toward the noise source. Check the manufacturer’s pickup pattern rather than guessing which side is the back.
Close an unnecessary door or window, quiet the fan only when thermally safe, and soften bare reflective surfaces with ordinary furnishings.
Set input gain so normal speech has healthy headroom and does not clip. Lower gain is not a cure if it makes you amplify the same noise later.
Prefer gain at the microphone, interface, or operating-system source when practical. OBS notes that its Gain filter is a fallback when the source cannot be adjusted earlier.
Open the filter chain for the correct microphone
Speak and watch the Audio Mixer to identify the microphone channel that actually moves. Similar device names make it surprisingly easy to process a webcam mic while recording a headset mic.
Use the cogwheel or menu beside that channel and choose Filters. OBS also allows source filters through the source’s context menu.
Give each filter a meaningful name such as “Gentle fan suppression” instead of collecting several indistinguishable copies.
Use the eye icon beside a filter to bypass and restore it. This is the most useful button in the entire exercise.
Start with one Noise Suppression filter
Add Noise Suppression and begin with RNNoise when it is available and your system has enough CPU headroom. OBS describes RNNoise as the higher-quality built-in choice with greater CPU cost. Speex is the configurable alternative; its Suppression Level becomes stronger as it moves farther below zero, and OBS explicitly warns that stronger settings can distort the wanted sound too.
Choose a method by listening, not by loyalty
RNNoise: a sensible first comparison for speech and mild environmental noise. Listen for softened consonants, watery tails, or changing ambience, and watch OBS performance under a real recording load.
Speex: useful when you want an adjustable suppression level or lighter processing. Start gently and move only until the noise becomes acceptable—not until the silence looks perfect.
Hardware-vendor methods: options such as NVIDIA Noise Removal require supported hardware and additional vendor components. Their presence and resource cost vary by operating system, driver, and installation.
No suppression: it remains a valid winner when the raw microphone is already clean or every processed version damages speech.
Decide whether quiet gaps need an expander or a gate
Suppression works on the noise content. An expander or gate responds mainly to level. An expander turns signals below its threshold down by a chosen ratio, so transitions can feel more natural. A gate behaves more like a door: below its closing condition the signal is strongly reduced, then it opens when speech crosses the opening condition.
Choose an Expander when you want breaths, keyboard noise, and room tone to recede rather than disappear. OBS recommends RMS detection for smoother threshold detection and documents a low-ratio Expander preset as a gentle starting point.
Choose a Noise Gate when you need the microphone decisively quiet between phrases and can tolerate a more audible transition.
Usually choose one, not both. Stacking a gate and expander can produce pumping or chopped word endings without adding meaningful control.
Neither tool cleans noise beneath your voice once speech is above the threshold. That is the boundary between level control and noise suppression.
Teach the threshold using your own room
Stay quiet in the posture and room you will actually use, and observe where the microphone meter settles. That is the noise floor for this session.
Speak softly, normally, and loudly from the same distance. The threshold needs room between ambient noise and the quietest speech you want to keep.
For an expander, raise the threshold until quiet noise recedes, then speak complete sentences and lower it if beginnings or endings become thin.
For a gate, place the Close Threshold above the noise floor and the Open Threshold higher than Close but below wanted speech. This hysteresis keeps the gate from fluttering around one boundary.
Tune attack, hold, and release by ear. Too-fast closure clips syllable endings; too-slow release leaves the room open long after a phrase.
Build a short chain you can still reason about
For spoken voice, a restrained chain might begin with Noise Suppression, use a Compressor only when vocal dynamics truly need control, place an Expander later, and finish with a Limiter. OBS’s filter documentation places the compressor near the beginning, the expander after compression or other effects but before a limiter, and the limiter last. Treat that as a defensible starting order, then verify the complete signal rather than turning it into dogma.
Noise Suppression: reduce mild steady noise before later dynamics processing reacts to it.
Compressor, if needed: reduce louder passages above its threshold. Output gain can also raise the remaining noise, so compare at similar perceived loudness.
Expander or Noise Gate, if needed: reduce quiet passages after the voice dynamics have settled. Use one unless you can name the separate job of each.
Limiter: keep it last as a final peak ceiling. It prevents excessive output peaks; it does not make a weak recording loud, remove noise, or repair clipping upstream.
Run a tiny blind audition before going live
Wear headphones and record the same 20–30 second passage with every filter bypassed. Include silence, soft speech, normal speech, a loud phrase, and one real noise such as typing.
Enable only Noise Suppression and record the identical passage from the same distance.
Add the expander or gate only if quiet gaps still need help; record again. Add dynamics processors one at a time rather than all at once.
Replay the saved files away from the live meter. Match playback loudness as closely as practical so “louder” is not mistaken for “cleaner.”
Listen for lost consonants, clipped first words, abrupt silence, pumping, metallic texture, breath exaggeration, and fan noise returning underneath speech.
Keep the least processed version that solves the distracting problem, then make a final test while OBS is encoding the actual scene you plan to use.
When the result still sounds wrong
Words lose their beginnings: lower the gate’s Open Threshold or shorten its attack; compare an expander for a gentler transition.
Word endings vanish: lengthen hold or release, or lower the closing threshold so natural decay remains audible.
Voice sounds metallic or underwater: reduce suppression strength, compare RNNoise with Speex, or solve more noise acoustically.
Silence is clean but fan noise returns during speech: the gate is behaving normally. Improve voice-to-noise ratio and tune suppression instead.
Noise becomes louder after compression: reduce makeup/output gain, revisit input gain, and compare the compressed and bypassed versions at similar loudness.
Audio crackles or OBS struggles under load: bypass filters one at a time and watch performance while recording the real scene; a method that exceeds available compute is not the right method for that machine.
You hear an echo while monitoring: check Advanced Audio Properties and make sure the microphone is not captured or monitored through two paths. Use headphones to prevent monitored audio returning through speakers.
A note for calls, streams, and platform processing
OBS filters affect the OBS source and its selected recording, stream, or monitoring routes. They do not automatically become a system-wide microphone for every meeting application. A conferencing service may also apply its own suppression, automatic gain, echo cancellation, and codec after OBS, so a locally clean recording can sound different to the audience.
Test the exact destination path, not only OBS monitoring.
Avoid stacking aggressive suppression in OBS and again in the destination application unless the second stage solves a measured problem.
If you use a virtual audio device, document the route and sample-rate choices so an operating-system update does not turn the next session into guesswork.
Keep a clean profile or duplicate scene collection before experimenting, especially for a production stream.
Continue shaping a reliable OBS setup
Set up OBS Studio on Ubuntu and verify a complete recording.
Install OBS Studio on Ubuntu from a maintainable package source.
Comments and corrections