I'm still not sure of how the audio is being generated. Is it something like this: vimeo.com/717517 ?

Hello kyle, sorry for the late response. I'll also document and share the source codes for this but for now I can say that I think it works quite differently from the example you posted. the mapping here is quite arbitrary but I think a bit of effective. I'm using several atari2600 sound chip emulators in parallel(every cell has one), when a cell becomes alive, it creates a "blip" but the parameters of that blip is determined by that cells history and position. Sound for the hotter cells last longer, for example. And the grid is also filtered from bottom to top. The bottom cells trigger lower frequency blips, and are bandpass filtered at lower frequencies, and the higher cells create higher frequency blips and filtered at higher frequencies. I think there were some additional subtle mappings, I need to check my code out.

Very cool! For some reason I wasn't considering the history of each cell as relevant.

Some portions still confuse me, e.g.: at 6:17 why does it sound irregular even though the pattern is repeating? Once it reaches a steady state, how does it manage to destabilize again?

Also, why start with a left-right symmetric pattern? Have you tried asymmetric patterns as well?

Actually the "history" thing was the highlight of this project for me. One of the purposes of taking history into account is to be able to achieve destabilization after stabilization. Each cell in this system keeps its history, and if a cell is "hotter" than or "cooler" than a specified value, the CA rules for that particular cell changes. When it comes into a normal range though, it starts to follow conway's rules again. When the system stabilizes, the remaining cells at rest gets very hot, while and remaining dead cells get very cold, so the entire system works on a different ruleset for a while, but when the hotness of the cells pass the extreme ends, they follow conway's rules again.

The irregularity you hear on repeating patterns are there because some of the synthesis parameters are controlled by pseudorandom number generators that have nothing to do with the ca system, they add their own dynamics.

That's very interesting -- I haven't heard of local modulations to automata rules before. It makes a lot more sense now. I'm kind of curious about more complex visualization involving "loop detection" -- where you actually compare the current state to previous states and see if the system is looping...

I think it would have been fun to hear the "pure" synthesis of the system, minus the additional randomness.