Universal Speech Enhancement with Score-based Diffusion
This work addresses the problem of universal speech enhancement for virtual communication and amateur recordings, proposing a novel generative approach that is incremental in its technical implementation.
The authors tackled the problem of holistic speech enhancement by addressing 55 different distortions simultaneously, and their system significantly outperformed state-of-the-art methods in subjective tests by expert listeners while achieving competitive objective scores with only 4-8 diffusion steps.
Removing background noise from speech audio has been the subject of considerable effort, especially in recent years due to the rise of virtual communication and amateur recordings. Yet background noise is not the only unpleasant disturbance that can prevent intelligibility: reverb, clipping, codec artifacts, problematic equalization, limited bandwidth, or inconsistent loudness are equally disturbing and ubiquitous. In this work, we propose to consider the task of speech enhancement as a holistic endeavor, and present a universal speech enhancement system that tackles 55 different distortions at the same time. Our approach consists of a generative model that employs score-based diffusion, together with a multi-resolution conditioning network that performs enhancement with mixture density networks. We show that this approach significantly outperforms the state of the art in a subjective test performed by expert listeners. We also show that it achieves competitive objective scores with just 4-8 diffusion steps, despite not considering any particular strategy for fast sampling. We hope that both our methodology and technical contributions encourage researchers and practitioners to adopt a universal approach to speech enhancement, possibly framing it as a generative task.