Testing Krisp, an AI noise canceller for video calls

Krisp installs on your com­puter, cre­ates a “vir­tual” micro­phone of the same name in the system and tries to remove back­ground noise during your Zoom calls. You can find all the details on the app’s web­site: krisp.ai (my refer­ral link).

I didn’t run the tests for the refer­rals, but since I’ve writ­ten the post anyway, why not add one :)

Use cases and results

I wanted to see how Krisp would work in my use case: a video call from a MacBook with AirPods in a noisy café. I emu­lated the café with var­i­ous record­ings from SoundCloud, played through the speak­ers of my home audio system. I recorded the sound in Loom, a screen­cast app. Zoom might behave dif­fer­ently, but it’s unlikely.

I’m happy with the result and will try it “in the field”. Given that there’s now a free plan with 120 min­utes a month, you can keep the app on your com­puter for calls from noisy places. In some cases it spoils the qual­ity of the voice, boosts the highs and dis­torts it. But in return it removes back­ground noise excel­lently and copes so-so with a voice nearby.

In every­day use, for the other person to under­stand me well, it’s better that my voice sounds worse but there are no dis­tract­ing voices or noise in the background.

That way the other person can turn up my voice and adjust, but they don’t have to “filter out” other sounds in their head. In my sub­jec­tive expe­ri­ence, voices are the hard­est thing to filter out in your head.

The second use case I want to test later is a call center, for its operators. I’m 90% sure it’ll be worse there, since the noise is mostly high-pitched and uneven: the neigh­bors’ voices. Most likely, for oper­a­tors the app won’t jus­tify its price.

A note on AirPods. AirPods Pro didn’t always con­nect prop­erly with Krisp and Loom together. I had to dis­con­nect them from the laptop and then con­nect them again. The ToothFairy app puts the device’s icon in the menu bar and makes this easier.

So, on to the tests.

Test 1: AKG headphones + loud music + an English-language show nearby

Variant 1.1: loud music + English-language show nearby + mic hang­ing loose, with­out Krisp

https://www.loom.com/share/a5ad779f5f4b4bddb7f5809f98675545

Variant 1.2: loud music + English-language show nearby + mic hang­ing loose, with Krisp

https://www.loom.com/share/72b51478947d431d94de7441c760a59c

Result: the volume is almost the same and the voice a little lower. The music is almost gone. The voice in the back­ground stayed; it got qui­eter, but it’s unpleas­antly distorted.

Score: 7/10

Test 2: AKG headphones + mic hanging loose + instrumental music, no voices in the background

Variant 2.1: instru­men­tal music, no voices in the back­ground + mic hang­ing loose, with­out Krisp

https://www.loom.com/share/86ba4d083e474e5ebbe1b3633196c944

Variant 2.2: instru­men­tal music, no voices in the back­ground + mic hang­ing loose, with Krisp

https://www.loom.com/share/87fa79d204174e9ea0880b2b0f7196b9

Result: the voice gets distorted. You can feel Krisp rais­ing its volume, and some­times the highs grate on the ear. The music comes through in snatches, but much quieter.

Score: 5/10

Test 3: AKG headphones + mic at my mouth + music with vocals + a video with loud Russian speech

Variant 3.1: with­out Krisp

https://www.loom.com/share/144e9f7cbb26429393117800935ec9ab

Variant 3.2: with Krisp (I forgot to raise the mic to my mouth)

https://www.loom.com/share/83be7536bc8048039db8b82dee6541c1

Test 4: AirPods + music with vocals + a video with loud Russian speech

Variant 4.1: with­out Krisp

https://www.loom.com/share/4754b6aa84ca487fbfd45ba42793c542

Variant 4.2: with Krisp

https://www.loom.com/share/5fefbe659e4748e280f50d8cc60de198

Result: for every­day use it’s better not to hear some­one else’s voice in the back­ground and to hear me qui­eter and dis­torted, than to hear me well with a voice get­ting in the way. The AirPods’ own noise sup­pres­sion is better than the wired AKGs’. The voice is qui­eter, but you hear less of the back­ground music even with­out Krisp.

Score: 7/10

Test 5: AirPods + café noise in the background, 70–80 dB

Variant 5.1: with­out Krisp

https://www.loom.com/share/edee5659a97645ca828111087bb3b056

Variant 5.2: with Krisp

https://www.loom.com/share/272e37ae5aec4ceab83eb8ebe7fae03e

Result: holy crap! It’s qui­eter, but the café noise is cut out com­pletely! Completely!

Score: 9/10

Test 6: AirPods + office noise in the background, 70–75 dB + Russian speech nearby

Variant 6.1: with­out Krisp

https://www.loom.com/share/3912d910ec084fa49fdbb60b28b0873c

Variant 6.2: with Krisp

https://www.loom.com/share/b11beff89ed94dcbb27d0576e6e0baf7

Result: it picks up the “neigh­bor’s” voice, and it dis­torted my voice a lot. You could call this the limit of usability. Sitting in such a noisy place is unpleas­ant in itself, so it’s not quite the every­day use case I wanted to test. Let’s bring the noise down to some­thing more usual.

Score: 3/10

Test 7: AirPods + office noise in the background, 60–65 dB + Russian speech nearby

Variant 7.1: with­out Krisp

https://www.loom.com/share/ca878d87226e4f548faf4e1591351c8c

Variant 7.2: with Krisp

https://www.loom.com/share/33e8724667cc42d29fe48d7fe111f53b

Result: a little of the voice is there, and over­all the noise sup­pres­sion is less noticeable. So it works best when the café is very noisy but there are few highs (nobody close by). In con­di­tions like these Krisp makes less sense, but it still makes some. I can hardly hear the “neigh­bor” and can’t hear the ambi­ent noise at all.

Score: 8/10

How I mea­sured the noise level in the room. I used the Sound Meter app from the Android app store. I think any sim­i­lar app will do if you want to repeat this.