AI speech synthesis has fundamentally made audiobooks accessible to any indie author — the real question isn't "is this possible?" anymore, it's "what does it actually cost, and what gets easily overlooked?"
The raw generation cost
For an average novel (roughly 80,000 words, ~480,000 characters, which works out to about 8 hours of audiobook length) with a high-quality speech model, raw API costs land around $48. If you instead sign up for a monthly plan with a common AI voice provider to get enough "character allowance" for a whole book, a suitable plan runs closer to $99 a month — noticeably more, because you're paying for unused capacity instead of just what the book actually generates.
What that number doesn't include
This is exactly the part many people underestimate: raw speech generation is only part of the work.
Chunking. Most services cap the amount of text per request at a few thousand characters. A chapter has to be broken into pieces — and settings (voice stability, expressiveness) need to stay identical across every piece, or you get audible seams within a single chapter.
Numbers and dates. Without preparation, pretty much every cheaper speech model reads years and large numbers unnaturally — "2025" becomes "two-zero-two-five" instead of "twenty twenty-five." Fixable, but not automatically — it takes an extra text-preparation step.
Character voices. By default, a single voice reads every line of dialogue in the same tone — hero, villain, child, all identical. Distinct character voices require your own mapping of who speaks which line.
Mastering. Consistent volume across several hours, no distracting background noise — raw speech generation doesn't deliver that automatically, it's a separate post-processing step.
Distribution. A finished audiobook still needs to be sold somewhere. Most distribution channels (Spotify, Apple, Google Play) are free to use (revenue share instead of a fee) and now explicitly accept AI voices — with one important exception: Audible/ACX still prohibits AI narration in the standard process.
Bottom line
Raw speech generation has become cheap and accessible — the real work sits in the steps around it, which rarely get counted when people talk about "cheap AI audiobooks." Anyone who walks through this themselves once quickly notices how many small decisions (model choice, voice settings, number handling, chapter assembly) sit between "upload text" and "finished, good-sounding audiobook."