Learning About The Pain In Acoustic Echo Reduction

After months of startup acceleration and incubation spent in our early stage, a few messages settle in your head. The one message above all is — do not build a product that solves a problem, build one that fixes somebody’s pain. Easy said, however really getting to know your potential customers and their pain is a time-consuming and energy-draining task. But an indispensable and rewarding one. And why do I care to write about that for a post in the category of technology?

It turns out retrospectively that we were using somewhat unconsciously the “lean startup” approach in the technical conception of our invention, which we believe fixes somebody’s pain as well :). Similar to asking potential customers for market exploration, our solution asks the very acoustic echo reduction (AER) system about its own pains. I know the last sounds a bit nerdish, but when you look at state-of-art solutions to detect double talk (crossings by zero, spectral coherence, likelihoods, etcetera), you start wondering which one of those really understands AER’s language. All? None?

Finding out AER pains is kind of checking whether your stomach can tolerate spicy Mexican food you have never tried before (but without the inconvenience of going to the hospital if something goes wrong). And our system is doing such a “meal tasting” as if it were for the first time, that is, every time a new type of spice. The ultimate challenge is to understand how well the system is digesting the exotic meal by listening to its reactions conveyed in its own language. Thanks to this particular interpretation of the lean approach, our technology for acoustic echo reduction is fully operative under any double-talk circumstance, be it noise, music, voice, or altogether. Towards true full-duplex hands-free voice communications.