LavaSR audio enhancer unveiled with real-time speech cleanup

LavaSR is a tiny but fast open-source model designed to clean up and enhance speech audio. At around 50 MB, it can run roughly 5,000 times faster than real time on a GPU and about 60 times faster than real time on a CPU, making it suitable for mobile and live use.

The model performs speech super-resolution and denoising, boosting clarity on noisy recordings without requiring heavy hardware. Public demos show LavaSR improving the intelligibility and presence of voice clips recorded in less-than-ideal conditions.

Users can try the model via a Hugging Face Space or Colab notebook and integrate it into voice apps, streaming setups, or editing workflows.