We have a very interesting mission that we believe in. The problem we are dealing with is the problem of human interface. When we speak, we are targeting some form of interface with humans or machines. And that's kind of the core of what we do and what we want to do. We want to make these communication more robust and immune to noise and interference.

So when you try to do speech enhancement, there are two metrics that we're trying to improve. One is quality of speech, how good does it sound. And the other one is intelligibility, did you understand it or not. In our work, we care about improving both of them at the same time.

And that kind of brings the problem of speech enhancement to a new dimension that's analogous to computer vision. In computer vision there is a very famous problem called image segmentation where the objective is to separate the different objects in the image. This is the car. This is a tree. This is a person. This is the road.

In speech, that problem has not been formulated like that before because it wasn't feasible to be done. With the advent of deep learning, that's possible today. And it's time to bring the deep learning technology to products in the field of speech.

And our first product is speech enhancement. We need to have a mechanism to deploy this from prototyping and research exploration to products in a short period of time. And bat365 Coder came to be very handy tool in that process. It enabled us to deploy this, convert it to streaming format, and deploy it into C for production in six man-weeks.