It is 9:14 a.m. somewhere. A leaf blower starts up across the street. A dog barks at the leaf blower. A toddler joins the dog. And on a call that has not paused for any of it, a sales lead in Manila explains pricing to a buyer in Atlanta. The buyer hears none of the chaos. That is Krisp, doing its day job.
Krisp is the voice AI company that decided audio quality was not a feature but a category. The desktop app strips out background noise, other voices, and echo before the sound ever leaves your machine. The SDK does the same job inside contact centers, BPOs, and call platforms most consumers will never log into. The numbers are showy - 75 billion minutes of audio a month, a reported $37.7 million in annual recurring revenue, integrations inside Discord - but the real product is simpler. Krisp sells the absence of things you did not want to hear.
"We want to build the voice productivity category, not a feature."
- Davit Baghdasaryan, Co-founder & CEO§ 01 / THE PROBLEMAn office is a luxury again
In 2017, "remote work" was still mostly a perk. Conference calls were a known annoyance and a tolerated one. Then 2020 happened, the world's spare bedrooms became its boardrooms, and the polite engineering compromise of "the listener can sort of figure it out" stopped being acceptable. Suddenly the background was the foreground - a barking dog, an espresso machine, a sibling on a Zoom call in the next room. The problem was not new. The audience was.
Krisp's founders had seen this coming from a strange angle. Davit Baghdasaryan and Artavazd Minasyan met inside Twilio, where Davit ran product security and Artavazd brought a mathematician's stubbornness. Stepan Sargsyan joined as chief scientist. Their thesis was almost annoyingly clean: the human ear has been doing real-time noise separation for a few hundred thousand years; a small neural network running on a laptop ought to be able to do something similar, on-device, with no perceptible latency. Everything else - the meetings, the translations, the call center deployments - flowed from that bet.
"Audio never leaves the user's machine. The model runs where the microphone is."
- Krisp engineering blog, on-device processing§ 02 / THE BETTwo ex-Twilio engineers walk into a city
The city was Yerevan. The company was originally called 2Hz - a wink to the lowest frequency humans can hear, and a hint that the founders were going to obsess over the parts of audio nobody else wanted. Armenia is not, in the usual sense, where you launch a voice AI startup; it has neither Silicon Valley's gravitational pull nor a stadium of cloud audio engineers. It does have, on a per-capita basis, an unusually deep bench in mathematics and signal processing. Krisp hired accordingly.
Funding followed, modestly at first. Sierra Ventures and Shanda led a $1.5M seed in 2018. Storm Ventures led a $5M Series A in 2020 as the pandemic turned every kitchen table into a recording studio. RTP Global extended the round by another $9M in February 2021. The total is small by SaaS-bubble standards - around $17M of disclosed funding - but the company has consistently traded headline raise size for product depth.
The File on Krisp
- Founded 2017, Yerevan / Berkeley
- Legal name Krisp Technologies, Inc.
- Founders Davit Baghdasaryan, Artavazd Minasyan, Stepan Sargsyan
- Team ~220 engineers, scientists, GTM
- Funding ~$17M disclosed across Seed and Series A
- Status Profitable trajectory; reported $37.7M ARR
§ 03 / THE PRODUCTSubtraction, then everything else
The original product was a button: turn it on, hear less. The model isolates the dominant voice and suppresses everything else - typing, traffic, the housemate's cooking podcast - in real time, on-device, with no cloud round-trip. It works on every conferencing app it can plug into, which is most of them: Zoom, Google Meet, Microsoft Teams, Slack Huddles, Webex, and so on. The original demand came from individuals. The serious money came when companies realized they wanted the same thing at scale.
Today Krisp does five things on top of that foundation. It transcribes the call. It writes the meeting notes. It pulls out action items. It translates speech to speech in real time through its AI Live Interpreter pipeline. And, in what is now its most-cited innovation, it converts accents - dynamically, mid-call, both ways. A Filipino agent's English can be softened to a US listener's expected cadence; a US caller's English can be smoothed for an agent in Manila to process more easily. Call centers buy this for the reason you might expect: their customer satisfaction numbers climb and their handle times drop.
"Our goal is to make every live conversation not only clearer, but smarter."
- Krisp company positioning, 2025The Krisp Tape
§ 04 / THE PROOFThe numbers behind the silence
A category is something you can point at on a balance sheet. Krisp's evidence is unusually concrete for a company whose product is, technically, the removal of sound. Its SDKs handle more than 75 billion minutes of audio every month - a number that sounds invented and is not. Discord embeds Krisp natively. TTEC, one of the largest customer-experience operators in the world, runs accent conversion across millions of conversations. Everise, Arrivia, Startek and a roster of CCaaS platforms route their agents' voices through it.
Krisp by the numbers
§ 05 / THE MISSIONVoice as the next interface, again
If you ask Davit Baghdasaryan what Krisp is for, he will not say noise cancellation. He will say voice productivity - a category he has been arguing into existence for years on podcasts and in a Substack he writes called the Voice AI Newsletter. The pitch is that the next eighteen to twenty-four months of voice AI will be less about generative novelty and more about live conversations becoming, as he likes to put it, "not only clearer, but smarter." Translation. Summarization. Agent assistance. Compliance. The same set of moves Salesforce did for sales pipelines, in real time, on voice.
The company's culture follows from this. Krisp is engineering-heavy and unusually research-led for a SaaS at its stage; the team publishes openly on deep learning for speech, hosts the Voice AI newsletter as a public artifact, and treats audio as a scientific problem rather than a UI problem. Distributed-first across Yerevan and Berkeley, with a chunk of the research team within walking distance of an actual observatory, the company has resisted the temptation to repackage itself as a Zoom plugin every time the venture market gets excited. It has been a voice AI company for almost a decade. It just took the world a while to notice.
"The mute button is dead. Long live the model."
- YesPress newsroom, on Krisp's category bet§ 06 / TOMORROWWhat changes if Krisp is right
The optimistic version of Krisp's future is that nobody talks about Krisp at all. Voice gets clearer. Accents stop being a barrier to who answers your support call. Real-time translation stops being a novelty and starts being a default. The interesting part is that none of this requires you to use Krisp's app; it just requires Krisp's SDK to be inside whatever platform you are already using. That is the quiet ambition. Be the layer, not the logo.
There are risks. Microsoft and NVIDIA both ship competent on-device noise suppression. Otter and Fireflies want the meeting notes market. Sanas wants accent conversion. The voice AI space is suddenly crowded with companies that read the same memo. Krisp's bet is that the company that has been doing this since 2017 - with the SDK volumes to prove it and the contact center contracts to compound on - has a head start that is harder to copy than it looks.
It is 9:14 a.m. somewhere. The leaf blower is still going. The dog has settled. The toddler is, mercifully, eating. The buyer in Atlanta has just said yes to a contract she did not realize was being held together, on the other end of the line, by a small Armenian-American company whose product is the things she did not have to hear. Krisp does not need her to know its name. It just needs to keep doing its day job.