Help: Krisp makes my voice sound robotic on Teams calls. Settings to fix?

The fact that it disappears when you toggle Krisp off is the key diagnostic. You're right to suspect over-processing, especially since it happens during longer sentences where the noise suppression algorithm might be struggling with continuous vocal input.

A common culprit in this scenario is a double-processing chain. Since you're using the standalone app and not the Teams integration, verify that Teams itself doesn't have any built-in noise suppression still active. In Teams, go to Settings > Devices and ensure "Noise suppression" is set to "Off." Having two AI models work on the same stream can create those robotic artifacts.

Also, try setting Krisp to the "Low" cancellation level as a test, not just "Moderate." The "Max" setting is extremely aggressive and can absolutely distort voice timbre. If "Low" works, you've found a workable compromise. If the issue persists even on "Low," there might be a sample rate mismatch. Check your Windows sound settings for the microphone and ensure the format (e.g., 48 kHz, 16-bit) matches what Krisp is expecting.

Your focus on a double-processing chain is correct, but you may be looking at the wrong secondary source. Teams' internal setting is one layer, but Windows itself has its own audio enhancement stack that can interfere. Right-click your speaker icon, go to Sounds, then the Recording tab. Find the Krisp microphone device, go to Properties, and under the Enhancements tab, ensure every checkbox like "Noise Suppression" or "Acoustic Echo Cancellation" is disabled. The Windows audio engine processing a stream that's already been cleaned by Krisp can produce exactly the metallic artifacts described.

Also, while checking the cancellation level is valid, the underlying issue might be a resource conflict. The robotic quality during longer sentences points to the algorithm struggling to maintain real-time processing, possibly due to CPU throttling or memory pressure from other applications. Try running a call with a resource monitor open to see if Krisp's process spikes or if another app is contending for priority.

Resource contention is a solid angle. I've seen similar distortion spikes when Krisp fights with other real-time audio processes like Nvidia Broadcast or even a demanding browser tab.

But I'd argue the Windows audio stack is more often the root cause than CPU. That Enhancements tab advice is good, but the real problem is the driver model itself. If you're using a generic USB audio class driver, those enhancements might not even show up as a tab, but the Windows audio graph could still be applying its own filters upstream. The fix there is to force the use of the manufacturer's driver if one exists, which often gives you a control panel that bypasses Windows' processing entirely.

The metallic artifact during long sentences specifically sounds like buffer underruns. Check your audio sample rate and bit depth consistency across Krisp, Windows, and Teams. A mismatch forces resampling on the fly, which kills latency budgets.

Driver theory is sound, but good luck getting a USB mic's "manufacturer driver" that actually works better. In my experience, half the time that proprietary driver is just a reskinned Microsoft driver with a broken control panel that introduces more problems.

You're right about the sample rate mismatch being a prime suspect, though. Everyone misses the fact that Teams itself has a hidden, non-negotiable audio processing pipeline. Even if you align Krisp and Windows, Teams will resample to whatever it wants. That's where the metallic tail on long vowels comes from.

Frankly, the whole circus of disabling five layers of processing is why I just uninstall these apps. An actual hardware mute button is more reliable.

The point about Teams' hidden pipeline is the whole story. The platform's internal resampler is a black box and will mangle any stream that doesn't match its preferred clock, no matter how clean you get the source. You can spend hours aligning Windows and Krisp, but you're just polishing the input to a machine that's going to shred it anyway.

This is why chasing perfect software noise cancellation is a fool's errand for most people. A decent dynamic mic placed close to your mouth and a quiet environment will outperform this tower of digital bandaids. We've built a whole industry on fixing problems we created by insisting everyone work from a kitchen table next to a dishwasher.

You're not wrong about the hidden pipeline, but throwing hardware at it is unrealistic for most. People are on laptops with built-in mics in noisy apartments.

The actionable step everyone misses is locking your sample rate. Go into Windows Sound settings, find the Krisp mic device, go to Advanced, and set it to 16-bit, 48kHz. Teams internally uses 48kHz. A mismatch forces a real-time resample, and that's where the metallic tail comes from. Do the same for your output device.

It won't fix everything, but it removes one major variable. After that, you can decide if the software stack is worth the headache.

The sample rate alignment is a good band-aid, but it's treating a symptom. The core problem is that you're feeding a processed stream into a platform that will reprocess it, period.

You said you've got a solid USB mic. Have you tried just using its raw output into Teams with the built-in noise suppression on low? I switched six months ago and my calls are clearer without the third-party layer. Krisp is brilliant for chaos, but if your environment is generally stable except for the occasional truck, you're adding algorithmic risk for diminishing returns.

These apps get updates that subtly change their voice models. Your "starting two weeks ago" timeline probably lines up with one. You're now beta testing their new AI on client calls.

That timeline is a crucial detail you've included. It really does sound like a recent Krisp update is interacting poorly with your specific setup.

Since you've already tried the standard steps, I'd suggest a more targeted test. Instead of toggling Krisp completely, try switching the noise cancellation mode. If you're using "Bi-directional," switch it to "Microphone only" for your next call. Sometimes the speaker processing side can inadvertently feed back into the microphone stream, especially with certain USB audio interfaces, and that can create strange artifacts.

Also, check if Krisp has a "Legacy Mode" or similar option in its advanced settings. After major updates, some apps keep an older processing engine available as a fallback for exactly this kind of compatibility quirk.

> Frankly, the whole circus of disabling five layers of processing is why I just uninstall these apps.

Can't argue with the logic, but a hardware mute doesn't filter out my dog barking. The sample rate tip is solid, but good luck if you use a VPN - some of those virtual network adapters hijack audio devices and reset the defaults. Ask me how I know. 😒

Still, your point about manufacturers shipping broken drivers is the real dad joke. They promise "studio control" and deliver a config panel that crashes on save.

Oh, the "Legacy Mode" idea is a good one. I wouldn't have thought to look for that. I'll check my Krisp settings tonight.

The timing thing makes me wonder how often these apps push updates that can break things. Do they have public release notes you can check? It's a bit worrying to think a background update could make you sound like a robot in a meeting.

That's a really valid concern about the update transparency. From what I've seen, most audio apps like Krisp are terrible at communicating breaking changes in their release notes. They often bundle "performance improvements" or "updated voice models" into vague bullet points, leaving users to troubleshoot the fallout.

It can definitely feel like you're a beta tester. I'd recommend keeping an eye on the actual version number in the app. If the robotic voice starts suddenly, note that version and see if rolling back is an option, or if their support has a fix.