Anryze Story from Anton Gera, founder, serial entrepreneur, CEO at Anryze.
Anton Gera, founder, serial entrepreneur, CEO at Anryze
Late one night in New York, me and my cofounder Mike were sitting in our empty office on 17th Street and looking over our P&L flow. As a call tracking and analysis service, Anryze was doing pretty well for a three month-old company. We had landed two enterprise clients and had eight more in the pipeline. We were growing naturally. And yet there was something concerning about that P&L Excel document.
That something was the huge expense for speech recognition services from Google and IBM. Since we were providing our own service for free as a trial for first two months and our clients’ needs for speech recognition were expanding rapidly, our costs were getting kind of scary.
Please note that we’re not criticising these big corporations, which have created incredible technologies. It’s just that at the time we thought — should things really be like this? Should we and other companies be dependent on the two or three global players that currently provide speech recognition services?
For us, the answer was obvious. We did not want that. So we came up with an ambitious idea to change the situation.
Right away we contacted our CTO and told him the idea. His reaction? ‘Oh, dude, not again!’
I should mention that we’ve been working together on speech recognition systems for the past six years and knew the state of play. A while back we even built our own SR engine. It was pretty good, but we were struggling with creating proper working educational models.
When we began to accelerate this idea, we met an ingenious mathematician, Sergey Levashkin Ph.D., who was working in New York as a data scientist on applied mathematical models for different companies which needed to implement neural networks into their businesses. Luckily, he wasn’t familiar with any speech recognition systems and so his mind wasn’t blinkered with existing solutions. He looked at our ideas and noticed that there might be a better way to acquire the necessary data from sound waves. That was a turning point for us.
As you can see, we and others trying to build new speech recognition engines were looking to the existing Google, IBM and Nuance solutions as the undisputed standard for how to go about it. This turned out to be totally wrong!
Our guys started researching Stanford and MIT studies and found several ideas that gave us a sense of how it should actually be implemented. Finally, we had found a whole new and even better way to create speech recognition neural networks! We were so inspired that we began hiring a new team right away.
So we put together the core of our future Dream Team: our management and operational guys, Anton, Mike and Max; our CTO, Oleg Zaychuk, who built his first speech recognition engine back in his university days, and who already knew all the open source solutions we would need to create our unique system; data scientist Sergey Levashkin, who brought an important fresh approach to data processing and neural networks into the company; and Oleg Fedosenko, a data scientist who had been working for several years on creating educational models for CMU Sphinx (an open source speech recognition engine).
The amazing news is that our first tests already give better results than IBM Watson — and keep in mind that these are only very early pre-alpha internal tests! I haven’t included details of the technology here because you can find it all detailed and explained at our White paper.
Our next post will explain more about Anryze, and why this technology will be so revolutionary.