STREAMING GUIDE / OBS STUDIO

Set up OBS microphone filters around your voice

Start with a clean capture, add only the processing you need, and listen for lost words as carefully as background noise.

START HERE

Fix microphone placement and input gain first. Record an unprocessed reference, then add one filter at a time. Suppression targets background noise, a gate controls pauses, compression reduces level swings, and a final limiter catches peaks. None can restore audio clipped before it reaches OBS.

Make an unprocessed reference recording

Disable existing OBS mic filters temporarily, noting their settings first. Use the same microphone distance and seating position you expect during the stream. Record quiet breathing, ordinary speech, a louder phrase, soft word endings, typing, and several seconds of room sound. Listen to the saved clip.

If normal speech is faint while room noise is loud, move the microphone closer and reduce avoidable noise before boosting gain. If loud words already sound harsh, reduce the microphone or interface input gain and retest. Turning down the OBS fader after distortion has entered the signal does not reconstruct the damaged waveform. The browser microphone test offers a separate input check, but browser processing may differ from OBS.

Choose the filter for the problem you hear

ProblemCandidateListen for
Steady fan noise under speechNoise SuppressionWatery or robotic consonants from excessive processing.
Unwanted room sound between phrasesNoise GateMissing first syllables and chopped sentence endings.
Shouts much louder than normal speechCompressorUnnatural pumping or room noise rising after words.
Occasional final peaksLimiterA voice held against the ceiling almost continuously.

Open the microphone source’s Filters and add only the next filter you can evaluate. You do not need every filter in the table. Keep a reference clip without processing so each change has something concrete to improve upon.

Use suppression gently; tune the gate from levels

OBS Noise Suppression is intended for modest background noise, not rescuing speech buried in a loud room. Compare a sentence with the filter enabled and disabled. If it removes breaths or turns speech metallic, reduce the effect where adjustable, try another supported method, or improve the microphone position. Additional suppression already running in the microphone’s own software may compound the change; if you use RTX voice effects, follow the NVIDIA Broadcast in OBS guide to avoid double processing. You can also inspect an exported WAV sample in the local WAV file inspector to check peak levels and quiet noise floors.

A Noise Gate opens for speech and closes during quiet passages. Set its thresholds using your actual quiet-room and soft-speech levels, not numbers copied from a different microphone. Keep the close threshold lower than the open threshold, and test the beginnings and endings of softly spoken words. If room noise and quiet speech overlap too much, placement may help more than a stricter gate.

Use compression to control variation

The Compressor reduces material above its threshold. Start with ordinary speech and then a louder phrase; choose a threshold and moderate ratio that affect the louder part without flattening everything. A ratio such as 2:1 or 3:1 is an experiment to compare, not a universal voice preset.

To lower the game or music when you talk, put a compressor on that source and use the mic to trigger it. See how to set up OBS audio ducking.

Adjust one control at a time. If the return to normal level sounds obvious, compare release settings using the same sentence. Add output gain only after checking the new peak level: making the compressed signal louder also raises its quieter content. If a compressor goes before a gate, retune the gate against that processed signal rather than its earlier readings.

Keep the final limiter as a ceiling

OBS recommends the Limiter as the last filter in the chain. Set its Threshold control below digital full scale and retest your loud phrase. For example, a Threshold of −2 dB provides some ceiling headroom at that filter’s output, but it is not a complete stream loudness target and does not protect against clipping earlier in the microphone or interface.

If ordinary speech constantly presses against the ceiling, return to gain and compression instead of asking the limiter to do all the work. Also check the combined program: several individually clean sources can make an overly loud mix when they play together. Lower competing game or alert levels rather than pushing the microphone indefinitely—see how to balance your mic, game, Discord, and alerts in OBS.

Compare a real minute, not just silence

Repeat the reference sequence with your game or call running. Listen for intelligible soft speech, natural pauses, and louder words that remain clear. Compare at similar listening loudness; a louder version can seem better even when it contains more artifacts. Bypass one filter at a time when the processed result is worse, so you can identify the stage responsible.

What is the best filter order? There is no single preset for every room and mic. A starting experiment is modest suppression, a carefully tuned gate only if pauses need it, compression when needed, and a final limiter. Putting the gate before compression keeps the compressor from raising room noise into the gate’s trigger range; retest any downstream threshold after changing an earlier stage, and remove filters that do not improve the recording.

Can filters fix wireless dropouts? Missing audio is a connection or capture problem. Return to an unprocessed baseline and troubleshoot the route before shaping its tone.

Once your filters sound clean, measure a short OBS test recording or live broadcast tab in the live stream loudness & LUFS checker to confirm your integrated loudness, true-peak ceiling, and limiter headroom.

Sources & version notes

Documentation checked November 18, 2026. Use the platform and version notes in this guide when comparing settings. The troubleshooting checks are intended for your own setup; this review date does not represent hardware testing.