codec2 VS jukebox

I'm currently working on a 2HD floppy disk release of a podcast episode which has taken me down an interesting rabbit hole of low-bitrate audio codecs, floppy disk aesthetics and label design.

The episode I'm releasing is 1h40m long, which sounds like it couldn't possibly fit in 1.44MB, but there's an obscure but free codec mostly used by digital HAM radio enthusiasts called Codec2 [1] which can encode human voice in as low as 700bit/s (roughly 315KB/hr). At that quality setting it sounds like complete garbage, although still barely intelligible, and there are no readily available media players for the it (there's not even an established file format) so I'm having to include complicated ffmpeg instructions in a README to get people to play it, but I think it all adds to the charm.

Funnily enough after searching for days for how to fit the original episode in a floppy (tried low-bitrate MP3, Opus and AAC, all of which fail at the task), and after discovering Codec2 almost by chance, I only then came across an article called "Codec2: a whole Podcast on a Floppy Disk" [2] which I probably would've found if I had started by searching "podcast in a floppy".

[1] https://github.com/drowe67/codec2

[2] https://auphonic.com/blog/2018/06/01/codec2-podcast-on-flopp...

1.5 kbps speech is impressive, I wonder how it compares with Codec2 https://github.com/drowe67/codec2

David Rowe has some information on doing this kind of testing here: https://github.com/drowe67/codec2/blob/master/README_fsk.md

You have a chance with codec2, but I wouldn't expect it to handle any real-time communication. This codec only works for voice, for anything else it's trash.

You also got some samples available at http://www.rowetel.com/codec2.html

openai/jukebox: Music Generation

https://github.com/openai/jukebox The demo code is there.

I didn't even know about it until I heard Sam Altman casually mention it in an interview, I was expecting some basic tunes generator, but this is so amazing! I mean yeah the voices are not clear, it's muffled, but look at how far have image models progressed, if you applied the same amount of collaborative effort here, the results could be amazing! ElevenLabs showed how good and clear can AI-created voices sound. The only reason I can think of is that the Jukebox code is under view license only.

OpenAI had this figured out 3 years ago: https://openai.com/blog/jukebox/ . You could then even define your own text. Model is open source too.

They've had jukebox for a few years now, so I'm sure some new model will get released and explode overnight, like what chatGPT did.

That already exists

This model is designed to output raw audio.

However, there are many models which do output midi. That's actually much simpler, and has been done already a few years ago.

I thought OpenAI did this. But then, I might misremember, because their Jukebox actually also seems to produce raw audio (https://openai.com/blog/jukebox/).

However, midi generation is so easy, you even find it in some tutorials: https://www.tensorflow.org/tutorials/audio/music_generation