It occurs to me that adding a visual watermark might actually serve to obscure a visual watermarking scheme that is otherwise invisible by providing data that scrambles or breaks the watermark decoder itself.
Audio watermarks can be distorted in any number of ways; and it could be that some of the wildly poor audio quality in most cam-rips is probably the only way you can defeat the watermark; by using a LQ microphone and encoding the audio to a very limited bitrate and then re-upsampling; to defeat any subtle alterations a digital watermark might make to the audio waveform.
Keybase is better than Signal. You may not like it’s current owners but it still works, still functions, and can be used to chat privately. It’s entirely OSS on the client side; and server-side software isn’t provided; but with an open Client; it’s likely trivial to reverse and re-implement your own. (Keybase itself doesn’t provide their server code; it’s private due to abuse constraints)
Keybase is End to End Encrypted. It may not be as “feature rich” but all features are private.
I’m not sure if it’s indev anymore though; and it does allow you to be as public or as private as you’d like to be about your identity.