LavaSR is a lightweight and high-quality speech enhancement model designed to improve low-quality audio inputs. It utilizes a 50MB bandwidth extension model combined with the UL-UNAS denoiser to restore and enhance speech effectively.
50MB model size for efficient deployment
Supports input rates from 8-48kHz, outputting at 48kHz
Inference speeds ranging from 10-50x real-time on CPU and 400-4000x real-time on GPU