Make your Teams calls sound studio-quality with one setting

It's 9:02 AM on a Tuesday. My golden retriever decides this is the exact moment to lose her mind at a UPS truck. I'm three sentences into a client call. The client says "wait — was that a dog or a fire alarm?" I want to die.

This was me about a year ago. I owned a perfectly fine USB headset. I just hadn't bothered to flip one toggle in Teams that would've made the dog basically inaudible. Took me three years of meeting embarrassments to figure out the setting was even there.

Teams has, I'd argue, the best real-time noise suppression of any major conferencing app. Better than Zoom's. Way better than whatever Webex is doing these days. And most people leave it on Auto, which is fine but not great. Bumping it to High before any call where you're not in a quiet room — and knowing when to dial it back — changes how you sound to everyone else.

Where to find the dropdown

Same path on Mac and Windows for the new Teams client:

Click your profile picture, top right.
Settings.
Devices in the left sidebar.
Scroll to Noise suppression.

Four options. Auto, High, Low, Off. If you're still on classic Teams (the purple icon — Microsoft is dragging that one out behind the barn through 2026) the menu is in the same place but the dropdown looks slightly different. Very old builds may not even show High. Update your client.

You can also change it mid-call, which is genuinely useful. Three-dot More menu in the call controls → Settings → Device settings. The same dropdown shows up in a side pane. Changes apply right away — no leaving the meeting, no audio dropout. I do this a lot when I move from my office to the kitchen.

Which mode, when

Mode	When to use it
Auto	Default. Teams listens for the first few seconds and decides. Decent for variable conditions.
High	Cafés. Open offices. Kids in the next room. Mechanical keyboards (especially those Cherry MX Blues people insist on). HVAC. Traffic. Anything ambient.
Low	Music or instruments you actually want others to hear. Light filtering.
Off	Recording, podcasting, music lessons, or when you've got an external audio interface that's already cleaning the signal upstream.

If you remember nothing else: High for noise, Off for music. Auto is fine but it's reactive — it samples the room at the start of the call and may not catch up if a leaf blower starts up across the street ten minutes in. Which has happened to me twice in the last month, because apparently every landscaping crew in my neighborhood operates on a synchronized schedule.

What's going on under the hood

This isn't a noise gate. It's not a high-pass filter. It's a deep neural network Microsoft trained on thousands of hours of speech and non-speech audio. The model runs locally on your machine, chews through audio in 10-millisecond frames, and classifies each frame as "speech" or "noise" before attenuating the noise bits while keeping the speech bits intact. They published a paper on it back in 2020 if you want to nerd out.

This is why High mode can vanish a keystroke entirely while your voice comes through clean. A regular noise gate would either let the keys through (gate too open) or chop the start of every sentence (gate too tight). The ML model dodges both because it's not making decisions based on volume. It's deciding based on what the audio sounds like.

A few practical fallout effects from this design:

Pre-recorded audio is treated weirdly. Play a YouTube clip into your mic and Teams will treat the speech in the clip as your voice and the music as noise. Usually fine. It's also why sharing system audio is a totally separate feature.
It can clip the very start of a word. Rare. Happens after a long silence sometimes — first phoneme gets misclassified as noise. If you notice it, drop to Auto or Low.
It needs a real signal. Whisper into a too-quiet mic and the model has nothing to work with. Boost your gain.

A note on hardware

If you're rocking an older laptop, High can spike your CPU. I've seen Teams' helper process climb to 22% on my friend's 2018 ThinkPad mid-call. On anything from the last four years — Apple Silicon, recent AMD, recent Intel — you won't notice. On Copilot+ PCs Microsoft offloads the model to the NPU entirely, which is genuinely cool the first time you see your CPU stay flat during a call.

If your fan kicks on during meetings and you suspect this is the culprit, drop to Auto and see. The cost is real but only matters on older silicon.

Test it, please

Settings → Devices → Make a test call. Bot records ten seconds and plays it back exactly the way the other side hears you. Try it once on Auto, once on High. The difference is usually obvious in the first sentence.

While you're testing, do these three things:

Tap your keyboard while talking. On High, the taps should be inaudible. If they're not, your gain is probably way too high.
Crinkle a sheet of paper near the mic. Should mostly disappear.
Hum a low note through a sentence. If your hum vanishes mid-syllable, the model's being too aggressive for your voice, and you'll sound better on Auto.

That last one's how I figured out my voice was being slightly chopped on High. Switched to Auto for client work, kept High for internal stand-ups where it doesn't matter as much. Different solution for different rooms.

Pitfalls that bite people

A handful of things I see go wrong constantly:

Bluetooth headsets in hands-free mode. When a Bluetooth headset acts as both mic and speaker, it falls into a junk codec called HFP that sounds like a 1997 office phone. No amount of noise suppression saves you from this — the audio is destroyed before Teams ever sees it. Switch your output to laptop speakers, or use a wired/USB headset. Yes, your AirPods. Yes, even the AirPods Max. Bluetooth mics are bad. I will die on this hill.
Two noise suppressors fighting. If your Jabra/Poly/Shokz headset has its own DSP-based suppression and you crank Teams to High on top, the two stack badly and you get a warbly underwater sound. Pick one. Teams' is usually better unless you have a high-end Jabra Engage or similar.
Wrong mic selected. Teams will sometimes default to the laptop's built-in array even with a headset plugged in. Check Settings → Devices → Microphone. The little volume meter should jump when you talk into the device you actually want to use.
Windows audio enhancements stacking on top. Settings → System → Sound → your mic → Audio enhancements. Turn this off if you're using Teams' suppression. Two pipelines compounding never sounds good.

Bonus tweaks in the same pane

While you're already in Devices:

Echo cancellation — leave on. This is what stops the other side from hearing themselves echo a half-second later when you have your laptop speakers up.
High fidelity music mode — only flip this on when you genuinely need to share music or instruments. It disables noise suppression and echo cancellation in exchange for full-bandwidth audio. Forgetting to turn it off after a music session is the classic foot-gun. I left it on for a week once and wondered why my calls sounded so rough.
Spatial audio — this is a listening setting, not a sending one. Doesn't change how you sound to others, only how others sound to you. Try it on a headset for big meetings — voices end up positioned around you instead of mashed in the middle.

If you present a lot, hunt down Voice isolation under Devices. Newer Microsoft 365 feature. It uses your enrolled voice profile to suppress other people's voices in the room around you. Game-changer in coworking spaces. Took me about four minutes to enroll my voice and I've never turned it off since.

So

Open Settings → Devices → Noise suppression. Flip from Auto to High. Make a ten-second test call to confirm you sound like yourself. Drop to Low or Off when sharing music. Turn off Windows-level audio enhancements so the two systems don't fight. With the right mic selected and echo cancellation on, most consumer headsets end up sounding like proper conferencing gear. No hardware upgrade. One toggle.

The dog still barks. But nobody hears her anymore. Worth it.