Abstract
We present a way to compactly represent acoustic transfer functions using a small, yet flexible, parametric representation. We show that we can use a neural network as a “soft” table lookup and train it to produce the value of transfer functions at arbitrary points in space and time. Doing so allows us to interpolate and produce unseen data, and to represent acoustic environments using a remarkably compact representation. Due to this representation being differentiable, this opens up multiple opportunities to employ such models within more sophisticated audio processing systems.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have