On the Design of Diffusion-based Neural Speech Codecs

Pietro Foti, Andreas Brendel

Recently, neural speech codecs (NSCs) trained as generative models have shown superior performance compared to conventional codecs at low bitrates. Although most state-of-the-art N…