I remember hearing stories in my childhood about how Armenian scientists moved mountains in Soviet times. After a while, I started to think that maybe those are typical stories from the series of “մերն ուրիշ է” (ours is different) that do not have much value other than being local legends. At last, I changed my views completely after seeing the real potential of Armenian scientists thanks to Krisp. 

This is what co-founder and CEO of Krisp Davit Baghdasaryan wrote on his Facebook page. Krisp is an AI startup that removes background noise from conference calls. It’s based in Armenia and has created a one-of-a-kind noise cancelling app. It has attracted attention from the tech world for its innovation, recording significant success and growth in a very short time (since 2017). Here are some numbers:

• Total $3.5 million investment raised from VC funds;

• 25 million cleaned minutes of noise in the calls of 30,000 users from more than 200 countries;

• recognized as the Audio & Video Product of the Year in 2018 by ProductHunt, a platform that presents the world’s latest technologies;

• featured on TechCrunch and other top media platforms that report on the business of technology, startups, venture capital funding and Silicon Valley;

• started cooperation with world-class companies such as Nvidia and Intel. 

These are impressive accomplishments for a two-year-old startup based in Armenia. But what’s all the “noise” created around this noise-removing technology and how can it serve as an example for other startups in deep-tech? 

I sat down with Artavazd Minasyan, the other co-founder, to get the inside scoop on startup life in a deep tech world dominated by a few major giants.

 

Science-based startups VS. regular startups

As Minasyan explains, regular startups have common growth stages: 

1. getting an idea/identifying the problem;

2. engaging a team; 

3. building a product; 

4. marketing the product; 

5. scaling the startup. 

The difference for science-based startups is that they have one more very important stage between the idea and the product stage: building new technology. The difference between building new technology and building a new product is scientific research. Engineering teams are not enough for science-based startups, which is why a team of scientists is needed from the initial stage. This team researches and creates new technology, which may later be applied to building a new product.

However, those scientific discoveries and innovations are not initially aimed at building a specific product. Take Krisp as an example. The team didn’t develop their audio technology specifically for the Krisp app. They took the problem of real-time noise cancelation for calls and started working on creating a technology that will remove noise from calls without decreasing audio quality. The product later became the real-world application of their technology.

 

How Did It Begin? 

Baghdasaryan used to work at Twilio, a San Francisco-based tech company with over 1000 employees. There, he constantly encountered the same issues during conference calls: the background noise and the busy video background during calls. He mentioned these in one of his talks with Minasyan, a long-time friend. 

Minasyan, a math Ph.D., was digging into machine learning and was looking for an interesting scientific problem to solve. Baghdasaryan’s idea of background noise cancelation was perfect. He invited Stepan Sargsyan, the current Chief Scientist at Krisp, to join the research and they all started the journey together. 

Both of them had a background in physics, a Ph.D. in math and several scientific publications between them but no experience in machine learning or DSP (digital signal processing). The latter is defined as “the process of analyzing and modifying a signal to optimize or improve its efficiency or performance” and is important for working in audio technologies. Hence, they undertook their first challenge and started learning ML and DSP and then researching and implementing scientific papers from the industry.

After six months of hard work, Minasyan got back to Baghdasaryan with the first prototype of the noise-cancellation technology. It left him speechless. Although it was not perfect and needed improvement, it was still a breakthrough in audio technology.

After nine years in Silicon Valley, Baghdasaryan left Twilio in 2017 and moved to Armenia. Minasyan had already assembled a small team and had the first prototype for their technology. They founded a company (formerly named 2Hz) with the goal of creating audio technologies that no one else could provide.

 

The Technology Behind Krisp

Krisp adds an additional software layer between a physical microphone/speaker and conferencing apps. Your voice gets through but background noise doesn’t.

As Baghdasaryan writes in his blogpost on Nvidia Developer blog, noise suppression means “suppressing the noise that goes from your background to the person you are having a call with, and the noise coming from their background to you.” It’s different from other noise-canceling technologies that are used mainly for hardware devices (headphones, laptops, microphones, etc.). They either create an opposite sound wave to neutralize noise (active noise-cancellation) or simply prevent noise from entering your ears just like sticking your fingers into your ears (passive noise-cancellation).

Other software applications haven’t yet succeeded in creating a quality technology that people will love because it's actually very complicated. Imagine how diverse real world noises are (babies, dogs, sirens, etc.). It’s difficult to invent a technology that recognizes all their variations. Krisp collected 20,000 noises, 10,000 distinct speakers and 2500 hours of audio to develop a neural network, which they called krispNet DNN. A neural network, according to Technopedia is “a technology built to simulate the activity of the human brain – specifically, pattern recognition and the passage of input through various layers of simulated neural connections.” Powered by a deep neural network with multiple “hidden layers,” their technology succeeded in recognizing and removing any type of noise, becoming the best solution in the industry.

 

The Business Model

Minasyan and Baghdasaryan initially designed a business model based on selling licenses for their technology to the big names in the industry: Skype, Zoom, WebEx and many others. However, they didn’t have the resources, experience and channels to reach them. Moving innovation out of academia and trying to market it is a unique challenge for science-based startups.

Next, there was the need to raise funds. Of course, every startup grapples with this but it’s a little different for a startup in deep tech. “We were creating complicated and new technology, which needed long-term research and improvement until it gets ready to be marketed and monetized. Hence, our only option was attracting Venture Capital funds to continue our work and grow our startup,” mentions Baghdasaryan. He also explains that, in order to work with VC funds, a startup needs to aim for a valuation of at least 1 billion USD. VC investments require a tenfold return. A startup may need to adjust its business model, technology and the company structure to meet such an aggressive future target.

Realizing this, Baghdasaryan and Minasyan started to question their business model and challenge their go-to-market strategy. Here, the idea of the Krisp application as a product was born. Instead of selling licenses of their technology to big players in the industry, they could become a big player themselves under the “Krisp” brand. This puts them two years ahead of the other large companies, which are still working on their own noise-cancelation solutions.

This business model helped them stay independent, attract investors, compete with big brands and lead the industry as they continue to test, find their market with real use-cases and target customers and improve their technology accordingly. Moreover, as Minasyan mentioned, they are two years ahead in the market and will try to leverage that with better quality and new solutions such as mobile versions of the app (iOS version is to be released soon), echo-removing, personalized noise cancelation (recognizing and keeping only your voice) and even video background removal, which was another challenge that Baghdasaryan suggested two years ago and which is now in the process of development.

Bridging the Gap Between Scientific Discovery and Commercialization

Such case studies prove Armenia’s scientific potential. What remains is bridging science with business to have more startups in deep tech.

Minasyan believes that if scientists get a chance to develop their own businesses from their inventions and research, they can create their deep technologies and boost the growth of science-based startups. This may require initial stage funding, which may be a big input for further growth. In the case of Krisp, the founders self-funded the research stage until they created the prototype of the technology, which then attracted VC funds. However, not all startups can afford to do that.

Another model is science incubation or acceleration in universities. They usually finance the successful projects developed in the scope of their programs. As a hub for networking, finances and other resources, they connect new scientific ideas with the business world. The Berkeley SkyDeck accelerator is an example of this. During the acceleration of Krisp, Minasyan and Baghdasaryan were challenged to question and change their business model by creating the Krisp application as the core product for their scientific innovation. They also grew connections, accessed a pool of investors and raised funds after successfully passing through the acceleration program. Krisp was the first Armenian startup at SkyDeck and became a success story that motivated others to apply. Five of those were accepted recently and are in the process of acceleration now.

This being said, funds and acceleration programs can become a key for promoting and growing science-based startups. Both of those resources are rare in Armenia, which is why there are few deep-tech startups in the local ecosystem. Krisp is breaking new ground in more ways than one.

also read