OBS Noise Suppression Not Working: Fixes for Clear Audio

How OBS noise filters work

In OBS, noise suppression is not a global setting. It is a filter that attaches to one audio source and processes whatever passes through it. If the filter sits on the wrong source, or the source's device changed, the settings can look perfect while the air stays noisy.

The filter offers several methods. RNNoise is the quality-first default, Speex trades some quality for lower CPU usage, and NVIDIA Noise Removal requires the NVIDIA Broadcast SDK redistributable before it even appears in the list. OBS documents the trade-off plainly, and its knowledge base adds an honest limitation: the filter handles mild background noise well, but it is not effective at large amounts of background noise.

Two more details keep triage quick. The filter list also contains a Noise Gate and an Expander, so one source can carry several noise tools at once and mask which one is doing the work; and filters are per-source rather than per-scene, so two microphones can behave differently inside the same scene.

Fix checklist, in order

Work through these before rebuilding your scene, and keep a bypassed reference recording of the raw microphone handy; comparing it against the filtered version is faster than trusting the live monitor.

If none of the checks changes anything, record two short clips, one with the filter bypassed and one with it enabled, and compare the files. The file comparison removes monitoring and sync from the equation.

1. Confirm the filter is attached to the right source and enabled. A filter on a second microphone or on desktop audio will never touch the main mic.
2. Pick the method deliberately. RNNoise for quality, Speex when CPU is tight; note which one you changed so the next test is comparable.
3. Set the level where the method exposes one. Speex carries an adjustable suppression level, while RNNoise runs with no settings of its own.
4. Check the filter order. Filters process in sequence, so a gain or compressor sitting above the suppression changes what the suppressor hears.
5. Remove a second processor. NVIDIA Broadcast, SteelSeries Sonar, or Discord suppression on the same microphone can fight the OBS filter or add artifacts.
6. Test with a recording, not the monitor. Record a short clip through the real output path and listen to it before trusting the meters.

Noise gate versus noise suppression

They solve different problems. A noise gate cuts audio below a threshold, which silences the background between phrases but also chops breaths and word tails when set aggressively. Noise suppression reduces a constant bed underneath speech. OBS lists both because they are complementary, not interchangeable, and using a gate alone will not quiet the fan noise that plays while you talk.

A common pairing is a suppressor for the constant bed and a gentle gate for the gaps. Set the gate's threshold just below the quietest speech, then check that breaths and word endings still survive before trusting it in a live take.

NVIDIA Noise Removal is missing

The NVIDIA method only appears after its redistributable package is installed, because the filter links against the Broadcast SDK. If the option is greyed out or absent, install the SDK from the OBS documentation and restart the app; the option appears only after that extra step completes cleanly. Hardware requirements come from the NVIDIA Broadcast guide.

Choose the suppression level with a reference clip

Speex exposes an adjustable suppression level, and moving it without a fixed reference turns tuning into guesswork. Record a fifteen-second reference clip first: one spoken sentence, one silent pause, and one loud moment. Then change one control per round and replay the same clip.

The clip answers the only question that matters: did the pause get quieter without the consonants getting duller? When the two move together, the level went too far, and the earlier setting was the better one. RNNoise carries no level of its own, so for that method the meaningful choices are the method itself and the filter order around it.

When to clean the archive instead of fighting the filter

There is a point where tuning costs more than cleaning. If the noise is transient-heavy, if the room changed mid-session, or if the recording already exists, another round of filter adjustments solves the wrong problem. Fighting a filter that OBS itself describes as limited with heavy noise wastes an evening.

A file pass on the archive solves the actual problem, which is the recording. Keep the filter work for live audio and hand the finished file to a cleanup pass.

For a series, that split also keeps the live settings stable: the filter chain stops changing once it works, and improvements land on the archive instead of on the broadcast.

When the recording is already made

Filters are live processors. A stream archive or a local recording keeps the audio it captured at the time, and adding a filter later cannot reach back into the file. If the archive carries a steady bed under your voice, the fix is a file-based cleanup. The same live-versus-file distinction runs through Discord's suppression failures.

Export and clean the recording

The rescue path for an archive is straightforward:

1. Export the cleanest track. A local recording with separate audio tracks beats a video-only archive of the stream.
2. Upload one file to SimpleClean. Up to 500 MB and 15 minutes; audio in WAV, MP3, M4A, AAC, FLAC, OGG, or OPUS, and video in MP4, MOV, MKV, WebM, or AVI.
3. Compare Original with Cleaned on a pause, a consonant-heavy sentence, and the loudest passage, then keep the better version.
4. Publish or re-upload the cleaned file and keep the original archive untouched.

What cleanup can and cannot fix

A stream recorded in a normal room usually carries a steady bed: PC fans, air conditioning, and room tone, with the microphone's own hiss underneath. That is what a light cleanup pass handles, and removing fan noise from audio online walks through the specifics. Keyboard clicks, mouse thumps, dog barks, and people talking at once are transient events, and reflections in a bare room need treatment rather than another filter.

The recording also carries a trace of the room itself. A small space with hard walls leaves a short reflection on every syllable, and no suppressor removes a room. Cleaning can lower the steady bed; treatment, meaning a rug, curtains, or a closer microphone, is the honest fix for the space.

A desk between you and the microphone adds its own share of thumps and reflections, and placement fixes those more cheaply than any processing pass ever will.

Prevention before the next stream

A few habits keep the audio chain predictable:

• Run one real-time noise processor at a time across OBS, the mic software, and the calling app.
• Keep the mic gain sensible for the distance you actually use.
• Record a local track for sessions you plan to publish, so a clean source exists later.
• Note the filter settings that worked, so the next rig rebuild starts from a known state.

Related guides

The live-versus-file distinction shows up everywhere: cleaning a video recording's audio walks through the video-side workflow, and the fan noise guide goes deeper on the most common steady bed in a streamer's room.