Subscribe
Sign in
Krisp Mac — Mute background noise during calls
Take calls from wherever you want without being embarrassed for the crying baby noise or airport noise in the background.
Krisp works with any conferencing app and with headset .
108
Replies
Best
@davit_baghdasaryan Is there any compression or degradation of sound quality with this? I'm wondering if it could be any use as a filter when recording vocals for music e.g. for small time musicians recording at home, limiting the need for soundproofing.. or for podcasters.
@joshuafry Krisp is designed to filter out everything that is not human voice. So it might not work that well for recording music.
@davit_baghdasaryan Having a "non-human voice filter" is defo very helpful. My other pain when using VoIP is the echo on meeting rooms. Is Krisp able to reduce the room echo/reverb?
@ricardo_a_muacho most VoIP solutions successfully address echo. There is only one situation where echo is still unsolved. Thats when there are multiple mics (laptops) in the same room and network latency is high.
What's more interesting is reverberation. Yes, Krisp will solve it soon. We are working on it.
This is really cool. I've been looking for something like this for some time now and was about to create it myself!
What's your tech stack and how long did it take to create?
@joey_tawadrous The tech is entirely powered by Machine Learning. It took us more than 2 years to come up with this DNN model.
This is perhaps a weird usage but would this improve the quality of my tutorial/course video's audio? It doesnt seem to be one of the applications of the tech?
@baadier that's right it will work. However please keep in mind that Krisp is currently better suited for call as it uses 16kHz sampling rate. For tutorial videos you might want to use higher sampling rate. 44kHz support is in our roadmap.
@baadier @davit_baghdasaryan Looking forward to 44kHz support for recording audio courses (& podcast episodes).
Seems to work quite well. Can sometimes hear a little bit of background conversation in behind the active speaker but only when they're talking. I wonder if the algorithm is affected positively or negatively by something like the beam-forming microphones in my AirPods?
Pros:Seems to work with my video and audio conferencing apps on Mac
Cons:No iOS app
Congratulations 2Hz team with the release. Impatiently waiting for the next meeting call to test it.
The demos are really amazing.
If it really works as promised, our customer success team will be happy to use it with the clients.
Do you have an ETA for iPhone or Android release?
@armen_mkrtchian would be awesome if your customer success team to tries it.
Unfortunately iPhone/Android are not in the roadmap yet. There are integration challenges on mobile architecture.
@davit_baghdasaryan First of all LOVE THIS! It's about time :).
I've been using the app for the past few days I've noticed voice sounds are a bit more round/muted. I'd imagine much of your work has centered around filtering out background noises by classifying between human voice/ background noise. Have you/do you do anything to enhance the signal coming come voice and add back sounds that make voice sounds sharper/more human?
Logical extension: have you ever thought about improving the clarity of speech? For example correcting things like poor connection, mumbling, heavy accents, etc.. I'm thinking something like a GAN to propose an even better version of the sound?
@daflatow This is a great observation. Of course we are aware of that problem.
In fact improving overall clarity of speech is our company's mission (2hz.ai).
We thought that background noise is the first thing we should start with and I think we did a great job there. Enhancing signal bandwidth is another thing we plan to deploy soon (16kHz->48khz). And finally making the signal sharper will be the ultimate improvement.
I must say that each one of these problems is really tough though.
Applying GAN sounds very interesting.
@davit_baghdasaryan Brilliant. Just read more on your site -- excellent approaches on all fronts. Last Q: Have you ever thought about utilizing an additional mic? For example your phone mic (when on computer) via companion app. When I thought on this problem that was the easiest thing I came up with. I'd imagine you'd get a lot of benefit from an additional signal in the same room one closer to the source you want to filter and one closer to the signal you want to enhance. If you do it can you pls call it binauribus? haha
@daflatow that's a neat idea. I think latency would be a problem though. Multi-mic algorithms are effective only when there is no latency
@davit_baghdasaryan yeah makes sense -- turns it more into an eng problem. Given background noise is relatively constant tho, might be a useful lagged variable. Also might be valuable for creating an offline "golden" data set for future training / experimenting. Anyway... super cool stuff. Thanks for taking my Qs!
This looks great. Looking forward to trying it on Monday.
Any thoughts on charging for it? Seems like a lot of effort went into this. I would hate to see it ‘go away’ in the future.
@scottw thanks for the great question. The Beta will remain free. We don't have any finalized scheme yet but I assume that it will be kind of freemium model in the future.
Hey Guys - congrats on your launch . Downloaded and enjoying the results very much thus far . I feel there is a slight loss of the original sound (but no more than some NC headphones I have used over the years) .. I take it this is just the 'phase flipped' NC method ? Or have you new wizardry ! Do share ;-)
Well done: 👍 “Your voice and any audio never leaves your device.
All the magic happens locally on your device. Krisp is designed with privacy in mind.”
Initial testing using Slack in a busy office gave fantastic results. Killer app idea.. and good use of AI to boot
Amazing product!! Using for conference calls is nice and all, but are you guys thinking of offering an API or SDK? Need that to improve Speech-To-Text accuracy in my app!
Definitely give a try
Pros:Awesome app, clears the noise during the calls, had been alpha user, the app constantly gets better each week.
Cons:There was high CPU usage, but the consumption got less during the last weeks
I'm a long-time believer in the idea that our phone numbers should be more like email addresses - permanent and free and accessible from any device wherever we are in the world.
Until now, background noise at home, on the shuttle and even in the office always destroyed the soft phone experience on desktop. I highly recommend that MacOS + Google Voice + Krisp stack.
The even cooler thing is that you can mute background noise not only on your own end, but also for the incoming communication.
Pros:This is just amazing tool!
Cons:Not available for mobile and windows :(
I imagine this super-super-super awesome product helping me not to be in panic when there is an important call and i cant find a calm place
@davit_baghdasaryan Just had a call the other day with a jackhammer in the background on my end. No one heard a thing. A+
One feature this made me think of: Allowing users to hear a sample of what their audio sounds like. With a new product it might be hard to believe what things can and can't be filtered out.
@daflatow thanks! This is in our roadmap. Could you elaborate how you would like to see UX wise? Would a "Test Krisp" in the menu make sense (giving before/after)? Or you would like to test it during the call?
I love this and it is so useful in shared office settings or when on a call on the go.
Pros:Very very useful and needed in call scenarios
Cons:None that I can see.
Why the app Krisp is opening a new network connection (and process!) every few seconds (and 1kB)? What is it sending to the internet? Looks like a no-no... 🚫 🙅♂️ :/ @davit_baghdasaryan
https://cl.ly/3b7113befe56
hey @ravbaker, we collect metrics information such as "how may mins have been processed", "how many have been noise-cancelled", "how many times the app has been closed", etc. It's every few seconds if you do things with the app (like click on it). If you keep it idle or just speak it won't be every few seconds. This data helps feeding back improvements to the product team backlog.
If you are concerned about audio - I, as the CEO, assure that Krisp doesn't and will never send audio data from your device (at least without your consent).
Also, agree that every few seconds is a stretch (bad impl). We will fix it as it also generates too much traffic on our backend site.
Krisp