I like how this sounds, but then again, I probably wouldn't

BoM could be as low as a sound card and a PC, or a Raspberry Pi (assuming a cassette recorder/player already exists). There are codecs like FreeDV which have a chance to be usable if you're looking for real-time playback on-the-fly (i.e. not buffering up a load of data over a longer time, and then playing out over a shorter time). But whether such codecs play nicely with tape (which will have slight speed and frequency deviations (wow/flutter) probably is for further study, since those codecs were not designed for tape. It won't sound very great (and they are more intended for speech, so the music will sound bad with such a codec), but perhaps if non-standard things were done (e.g. record at a higher tape speed, or use many tracks on the tape, etc., then that could help perhaps.

Rewind to the 80's  when magnetic tape was being used for all sorts of things but usually with some specialist hardware to bypass the consumer analogue inputs/outputs and without the modern digital compression schemes.

Philips tried to market a 'Digital Compact Cassette' device commercially in the early 1990's, presumably hoping that the backward compatibility with analogue cassette tapes would give them an advantage over MiniDisc but both appeared to lose out to DAT.

https://en.wikipedia.org/wiki/Digital_Compact_Cassette

I recall that there were some lower end multitrack recorders on the market (late 80's) that used all four tracks of the standard compact cassette simultaneously (no tape flipping) allowing you to play tracks back whilst recording to another, to build up a mix old school. If you can get hold of one of them then you could perhaps increase your data throughput.  

Minidisc got me through a couple of years of post-uni study : ) For various reasons, I didn't manage to turn up to lectures often, so I purchased two devices, and a friend would record the lectures and snail-mail them to me every few days. Was quite fun, my friends would whisper gossip to me, interspersed with the lectures, so that I wouldn't feel like I was missing out. Minidisc was really excellent technology as far as I was concerned. And they found a few niches e.g. the smaller radio stations would use them for recording programs, since they were relatively low-cost and had good-enough audio.

But then MP3 players came along too quickly, everyone wanted the iPod.

I had access to a rackmount MiniDisc recorder for a while but by that stage I already had access to solid state audio field recorders, so it was more of a historical curiosity by then, even though they were still being sold up until around 2013.

stanto said:

Then have write it to a cassette tape, like in the ZX Spectrum / Commodore 64 days.

The likes of the BBC micro used the 'Computer Users Tape Standard' (CUTS) to record to cassette which appears to have been around 1200baud. 

https://en.wikipedia.org/wiki/Kansas_City_standard#Enhancements

Not sure how the ZX Spectrum / Commodore compared, but I don't recall them being any faster to load.

You could start with something simple like MIDI.

CUTS used two tones, but you could use more.

If DTMF had been faster, then perhaps that could have been an interesting approach.

I did some experiments on digital data storage on cassettes a couple of years ago. Whatever the kind of data you want to store (compressed music, raw samples, etc) keep in mind that you are going to have a VERY low bitrate.

You can look into how digital data is normally stored in cassettes, and you'll find that they pick a frequency to represent "1" and one to represent "0". You would normally read at least 4 cycles at the slowest frequency to detect a one or zero, so your "baud rate" will be limited by how fast it would be to read 4 cycles of the slowest frequency you select.

Due to the frequency response of the magnetic substrate, a tape can't record anything above 16 KHz or thereabouts. And at that limit you are already getting a severely attenuated wave that will be hard to tell apart from noise. There are fancier, more expensive tapes with better frequency response, but the average tape was very limited.  So let's say you select 12KHz for "1", and 6 KHz for "zero" (which is already kinda pushing it).

Because you need 4 cycles of the lowest frequency (traditionally at least; you could try to get away with using less cycles), your highest baud rate in this scenario would be 6000/4 = 1500 bits per second. And that is without considering adding error correcting bits, or start/stop bits, etc.

Most computers from the 80s that used tapes would use lower frequencies (1.2KHz or 2.4 KHz at most, probably for reliability and to avoid getting close to the point where the tape starts attenuating the wave) and would end up with a baud rate of like 300 bits per second. 600 baud was fast, and 1200 baud was pretty much "turbo".

If I were you, I'd start by trying to figure out a format/compression that would allow me to represent a song at those low bitrates and check how that sounds, while finding a way to convert "normal" songs to that.

Here is an example that indicates what hardware might work:

https://zeninstruments.blogspot.com/2021/10/manchester-decoder-and-cassette.html

As an aside: I always wondered if a modem signal that would be used on an old analog telephone line (POTS) could be recorded on tape. Those old modems used some pretty interesting schemes to increase the baud rate, even with bandwidth limited analog audio phone lines. 

There is a good article on theis subject (almost) on Wikipedia:

https://en.wikipedia.org/wiki/Modem

It explains how you can get up to 56000 bits per second over a telephone quality audio channel.

I don't think that anyone ever optimised a codec for audio cassette data recording but since you can get 12kHz audio bandwidth and better than 45dB signal to noise ratio I would guess that  better than 100kbit/s is possible.

There is an article here

https://electronics.stackexchange.com/questions/596747/how-much-data-could-be-stored-on-a-standard-compact-cassette-using-modern-encodi

which discusses the subject froma  different point of view and concludes a (probably) lower rate.

MK