In 2018, a company I was working for asked me to develop an open source solution for audio watermarking. At that point, we didn’t even find a single open source software that would be close to being usable in production.
As a result, today I am making the source code of “audiowmark” publically available under GNU GPL3 or later. It has many features, it is robust, fast, secure and of course we believe that the watermark is not audible for most users.
More infos, source code and audio demos are available on the audiowmark web page.
2 comments ↓
Hi Stefan, I am recently working on a watermark project too. I would like to find an algorithm that can survive under second-screen(audio being recorded). Do you have any recommendations? I tested your algorithm, it seems it can’t survive this scenario.
Hi! I think the audiowmark algorithm itself could work, it is just a question of getting the parameters right. One thing to keep in mind is that if you record, there is only one channel of audio left that carries information, whereas for the typical case (audio transcoded to lossy codec), there are two channels that carry information. Also playback/recording may damage the watermark even more. So I’d expect that you would have to use a higher watermarking strength for this to work (https://uplex.de/audiowmark/README.html#strength). This will reduce quality but improve the chances of the watermark to survive.
I just did a small experiment where I used strength 20, recorded the audio with my smartphone, and the watermark survived recording. Note that it only seems to work if I specify –strength on “audiowmark add”, but not on “audiowmark get” (unlike the documentation recommends). This may just mean that the documentation recommendation is not so good in this case. Systematic experiments would be necessary to determine which strength is best for your use case.
Leave a Comment