I was reviewing a customer call recording yesterday, and my dog decided it was the perfect time to alert me to a squirrel outside. The bark was loud and sudden—right in the middle of the client speaking.
I use Krisp's noise cancellation on my daily setup. Out of curiosity, I checked the raw microphone audio versus the Krisp-processed version from the call. The difference was significant. In the processed audio, the sharpest part of the bark was removed, but a lower, rumbly growl and some whine came through. I'd estimate it caught about 80% of the noise.
This got me thinking about the edge cases. Has anyone else done similar real-world tests with Krisp? I'm particularly interested in non-steady-state noises like:
* Sudden keyboard clatter (mechanical switches)
* Doorbells or phone ringtones
* Background conversation that isn't constant
I rely on clean audio for customer success check-ins, and knowing exactly what might slip through would help me manage my environment better. The documentation talks about "background noise," but it's light on specifics for these intermittent spikes.
That's a cool real-world test! I never thought to actually compare the raw and processed audio files like that. The "rumbly growl" part getting through is interesting. It makes me wonder if Krisp is tuned more for higher-frequency spikes and maybe struggles with lower, sustained vibrations, even if they're sudden.
I've had it mostly cancel out my mechanical keyboard, but that's a more consistent clatter. Your point about intermittent background conversation is what I'm curious about too. If someone walks into my home office and asks a quick question, does that get filtered as "background" or treated as a primary voice? The docs aren't clear on that logic.
Have you tried testing it with, like, a phone notification ping? That's my next worry for my own setup.
rookie