Hello,

Is it possible to add hidden text to a mp3 file, for example a nickname from somebody to identify from who the mp3 came from?

Unless I'm mistaken, you should be able to do that by writing into the "Comment" field of the ID3 tag.

comment can be deleted in an easy way, I want the text to be non-deletable (maybe write it in the header?)

If you mean hiding text other than in metadata (tags) ... like steganography or watermarking ... a similar question was asked recently :

http://www.hydrogenaudio.org/forums/index....showtopic=38137

related: http://www.hydrogenaudio.org/forums/index....showtopic=38065

There are places in the mp3 stream (directly in mp3 frames) where you can store some things without altering the audio data. But it's easy to spot this by an attacker and easy to remove. The only somewhat safe thing would be a place which affects the audio data. If done well, an attacker can not check whether there has been inserted something (hidden) or not and he/she can't destroy those informations without massivly altering the sound data again.

The Fraunhofer (IIS) people developed a watermarking scheme which CAN add a watermark directly into an AAC stream (affecting the audio data) in such a way that the watermark can be even detected in the decoded PCM version if you know the key that has been used for embedding that watermark. (That should work for MP3 as well.)

This is probably what you want (and the only thing which I'd consider safe enough) if you're going to sell stuff on the internet or whatever I guess the Fraunhofer people would love to talk to you in that case.

Sebi

edit: typos & grammar

Hmm... i'd be interested in knowing what krikke had in mind.

In any case, I was curious to know if a watermark like this would survive being encoded, transcoded, etc. by lossless encoders. A good watermark would obviously have to not perceptibly affect the sound its watermarking, so an encoder should be able to strip it out or modify it if the only criteria for encoding is maintaining PERCEPTUAL transparency, right?

Google for "mp3 ancillary data"

That's a common misconception. It also depends on the watermark's robustness. Usually more robust goes with transporting fewer bits. IIRC the Fraunhofer waterkarking sheme only distributes effectivly 150 bits of information over one second of audio using (possibly) heavy redundancy/error correcting codes. The quantization noise an encoder produces barely harms such a robust watermark.

Sebi

This is my old idea, to perfom the perfectly "transparent" steganography without any padding or ancillary".

The quantized values of MP3 are coded by pre-determined Huffman codebook. Endoders can select the "one" codebook to encode the values. Usually, encoder selects the BEST one to minimize the bitrate.

But, sometimes, there're two or more "BEST" codebooks. You can hide the information by which codebook to use when such a situation.

Hacky but it works for many audio/video coding system