search

PecholaL / IDEAW

Robust Neural Audio Watermarking with Invertible Dual-Embedding
Apache License 2.0
17 stars 1 forks source link

Request for Guidance on Robustness Testing in IDEAW Framework #3

Open xiaowenguo-hub opened 4 days ago

xiaowenguo-hub commented 4 days ago

Hello, I am currently studying your work on the IDEAW framework, and I find it very insightful for audio watermarking research. I have been testing the provided implementation for embedding and extracting watermarks, and I aim to evaluate the robustness of the system under various audio attacks. Could you kindly provide guidance or share any sample code to perform robustness testing for the IDEAW framework? Thank you

PecholaL commented 4 days ago

Hi, thanks for your interest in IDEAW. As the previous tests were performed on an internal company server, I currently do not have the robustness evaluation code readily available to me. The robustness evaluation code is generally written following this process: audio+msg+l.code -(embedder)-> watermarked audio -(attack (e.g. attack in attack layer))-> attacked audio -(extractor)-> msg+l.code, then calculate the ACC of the extracted msg+l.code is extracted. Note that the applied attack can be any one of the attacks in the attack layer or a combination of multiple attacks, and the parameters of the attack can be different from those in the attack layer.