I'm not sure about how much experience you have in writting software, but if what you want is to hide some data information within an analog audio signal, it seems for me like the way to go would be to use spread spectrum modulation and encode the data digitally and add it to your audio signal as some background noise.
Spread spectrum is in itself already a form of encryption, so the bytes you add there will be from the get go, very difficult to recover for a hacker unless they: 1-know that are there, and 2-have the key for generating the pseudorandom sequence to de-spread the signal.
But if you want even more security you can choose to packet your data by SSH tcp packets, and have in your phone a program to extract those from the digital signal.
As for point number 2. My suggestion immplies that you add something to the analog signal, or if you will to the PCM raw waveform, which is the same for all practical purposes. So, taking into account that the job of an encoder/compressor like mp3 is to compres the signal, but to eventually recover the original waveform with as less RMS error as possible, we could say that the original noise waveform that you add will still be there, slightly modified, but still there, and hence when you recover the bits by demodulation, and apply suitable Error correction tecniques, I think your data should still be there intact as it was put into the sound.
Please take into account that i just suggested a possible technology, normally used with narrowband RF signals, but maybe you could do some research and check if it will work in your scenario.
Also Another technique that could be used, if you want to send Text (Like they do in the video), is to generate based on your text a signal that will look like the ASCII characters in the time-frequency plane. Then to recover you just perform time-frequency analisys and use any image recognition algorithm.
This just came to mind, so it probably sounds very wierd, but may be another way to add another signal to an audio file.
Hope this helps.