A Full Cast AI Audio System is in the works --

Cross-posted from AI Marketing for Storytellers.

I’m back on my bullshit, baybee.

I was depressed a good chunk of last week, possibly because I wasn’t using AI to shove a boulder up a mountain — my body sensed the freedom, and was immediately “mooooore weeeeight” lol.

But then we realized the next highest value most difficult thing we could do was to somehow, some way, use the Arachne method — recursive, iterative, AI processing — to create full cast audiobooks, and I was off to the races with my trusty frens.

I think my first piece of advice here is to start a Youtube channel immediately.

It takes awhile to get monetized, so if you already have audiobooks, you’re gonna want to get on that.

If you’re thinking, “But I need my books to be PG-13 to fly!” — Momma already made an app for that, silly — Unspice.ai — get thee hither and check it out. (Substack subscribers get special love, as always — dm me your preferred email for Beta access!)

(And I actually filmed a demo of me using Arachne and Unspice both, here — but the offer mentioned inside of it, is only good for people who’re attending the Expand summit, coming up, and/or Substack Members.)

My Youtube’s been monetized for ages but everything on it’s been slow lately because I haven’t had any new material to give to them.

But soon I’ll have full cast AI audio!

This is the sum total of monies I have made on my Youtube, ever since its inception in 2022 — you can see where I dropped my original dragon human narrated audiobooks (after laboriously editing them by hand, ugh!) and then more dragon human narrated audiobooks, and then my Nightshifted series (which were entirely AI, me burning off 11labs credits.)

Amortized over time, it’s not a lot?

But it’s also not a little — and it really is passive income, in all senses of the word.

Only — you’ve gotta make the math math.

And doing work with 11labs credits is NOT it. (Also, not Spoken, either, $300 an audiobook? Fuck that.)

Because I’ve figured out a system that’s pretty wicked, using one new AI audio system and this time four — four! lol — fighting AI back end systems to make it work.

I feel like a supergenius again, and the world inside my brain has righted itself.

And here’s the evidence — a seven minute stretch with five AI voices: Rhaim, Sibyi, Helkin, Finx, and Jaegar, all parsed and pulled together, entirely without me.

I don’t have chapter markers wired in yet, so that’s why the scene seems to jump at the end, but that’s OK, that’s for post-post processing, lol.

It’s actually been a fascinating process, figuring out a) how to do this and b) cheaply as possible — we’re doing a full Bend Her book run now, from scratch, to get the final tallies, but I can tell you I haven’t spent more than $300 total on this project, and that’s with extensive testing and several dead ends.

It was, like everything else AI-related, torture for the first three days, and then weirdly fun — for instance, did you know that ffmpeg can read audio into waveforms and detect gaps that shouldn’t be there? — and that there are huge lists of homophones online, and word frequency calculators, etc etc etc?

So much data, and so much of it helpfully deterministic!

Then, on top of that, has been this whole delightful data management process — keeping track of everything on the back end — PLUS borrowed liberally from Content Factory’s sauce (which I’ll put below for newcomers) — in that I have several different lanes running slightly asynchronously working on this for me, in parallel.

And I, uh, need to show you all how I used Content Factory to localize 28 ebook covers, no? Heh! I think I forgot to do that!

It is the best kind of nerdy, and I’ve gotten to feel like Sherlock Holmes and Q all rolled into one.

Part of the reason this has had to be such a serious endeavor is that the end goal is doing this in 28 languages for each book, none of which I speak, so once again I’ve managed to corner my fool self into doing the impossible, because I’m making myself do it twenty-eight-bounds at a time.

My plan is to listen to each of the final books in English as they’re created. I’ll post them everywhere I can unabridged — including my own store! — and then I’ll run them through Unspice to post them on Youtube, then branch outwards in careful circles, starting with my most potentially profitable languages along the way.

I’ll be able to then do a diff between Unspiced full cast English audiobooks and German audiobooks, so that we can make sure we catch everything, roughly, muhahaha, bc I’ll have a full, solid, dataset.

And then alongside all this, I’m also automating a system that’ll turn my mp3s into mp4s for said Youtubery — I’ve already got a branch of the content factory churning out backgrounds — and that’s forcing me to get way more serious on that particular automation-ing front.

I was reminiscing with Novae — and looking at the full back end — this project is about 2x as gated (ie: the parts where it gets looked at, it gets filtered a jillion times harder, because it’s easier to gate, frankly) as the original Project Arachne translation machine, but it didn’t cost me another several thousand hours of my life, because it was already built on Arachne.

Plus I’d already taken a few days off back in March trying to create this thing with the AIs available to me back then (4.6 & 5.4? and 11labs only) so that I could make a voice cloned version of my How to Think with AI book.

Which kind of proves several of my chapters in there, heh — that everything you do with AI builds and compounds — even if it looks like a lot of starting-stopping-zigging-and-zagging along the way! <3

It’s never too late to learn, or too early to start.

xo!

Cassie

PS — this past week’s zoom is at the link beneath the paywall below, plus Content Factory, just ask for permission to access it! <3