M4869/WavMark · Apply for community grant: Academic project (gpu)

I am an intern from Microsoft Research Asia. We have recently proposed an AI-based audio watermarking model named "WavMark" to address the growing misuse risks of audio generative models:

WavMark: Watermarking for Audio Generation
https://arxiv.org/pdf/2308.12770

This model can encode up to 32-bit of watermark within a mere 1-second audio snippet. The watermark is imperceptible to the human, difficult to remove, and robust against ten common types of attacks.
Moreover, compared to the existing open-sourced watermarking tool, it achieves up to a 28-fold improvement in robustness.

We have recently open-sourced a toolkit based on it: https://github.com/wavmark/wavmark

And we would like to deploy an online demo on Huggingface using T4 GPU to showcase its capacity.

We also provide a demo of the watermarked audio: https://aka.ms/wavmark