Results after a month: Did Krisp actually improve my call feedback scores?

The confounding variable of your new USB mic makes isolation tricky. Even with a basic model, its directional pickup and likely higher bitrate provide a cleaner source signal before Krisp ever processes it. The real test would be running the USB mic with Krisp off for a week, then on for a week, while keeping all other factors constant.

You could approximate an A/B test by analyzing the spectral output. Record a sample with your keyboard typing using the USB mic alone, then with Krisp enabled. A tool like Audacity would show you the noise floor reduction. The mechanical keyboard is a consistent, intermittent noise source, which is what Krisp's model is trained to target.

I suspect you're seeing a compound effect. The mic improved your baseline clarity, and Krisp's noise gating removed the distracting peaks that cause listener fatigue. The plateau likely depends on your ambient environment. If the keyboard is your primary noise, the benefit is consistent. If you introduce variable noises like construction or pets, that's a different stress test for the model.

That's a great result to see. I think you've hit on a common challenge with these tools - isolating the variable is really tough. Even without a formal A/B test, your scenario points to the most likely answer.

You've got a classic two-factor improvement: a better source signal (the USB mic) and a noise gate on the output (Krisp). The mechanical keyboard is a perfect use case for Krisp's model. While the new mic might pick up less ambient sound directionally, those key clicks are loud and proximate. Krisp is almost certainly stripping those out, which directly impacts listener fatigue during demos.

That subconscious attention you mentioned is real. Investing in the mic probably made you more aware of your environment, leading to other small improvements like closing a window or muting when not speaking. It's a positive feedback loop. I'd be curious if the scores stabilize now that the new setup is routine, or if the clarity becomes the new consistent baseline.

That's a good point about the subconscious attention. I started using a noise tool and it made me more aware of my room's echo, so I added a rug. It's hard to untangle.

Did you check if your call scores improved more on days with longer typing, like during demos? That might hint at Krisp's direct impact versus the mic just being better overall.

Great idea about checking if scores improved more on heavy typing days. That's a clever way to triangulate the data.

It reminds me of a similar thing we tracked in our support team. We found customer satisfaction scores dipped slightly on days with lots of background dog barking, even if the agent was clear. Once we started using noise suppression, those dips smoothed out. It wasn't the main mic quality, it was the specific noise event.

You might be onto something. If the score bump is mainly from the new mic, it should be consistent. If Krisp is doing the heavy lifting, you'd expect a bigger positive delta on days with more keyboard activity, like during a live demo or troubleshooting session.

Yeah, but your support team's data is exactly the problem. Correlation is easy, causation is hard.

Scores dipped on days with "lots of background dog barking". Did you control for *other* factors on those days? Maybe agents were more stressed, calls were longer, or the issues were tougher. Noise suppression smoothed it out, sure, but you're assuming the noise was the *only* variable. Classic vendor-friendly interpretation 😏

Same goes for keyboard activity. If someone's doing a live demo, they're probably more engaged and clear anyway. The typing might just be a marker for a better presentation, not the thing ruining it.