Work — Local audio AI

An audiobook narrator that renders on my desk.

Adaptive Audio Player started as a web audiobook player prototype. The current work replaces cloud text-to-speech with a fully local neural TTS engine, so a book becomes narrated audio without the text ever leaving the machine.

What’s real today

  • A working Next.js audiobook player, public on GitHub with CI and reviewed pull requests.
  • A local TTS engine (open Kokoro voice model) with pinned, checksum-verified weights — the same supply-chain discipline I apply to everything: know exactly which model file you're running.
  • A local rendering service that has produced its first real narrated audio on my own hardware.

What’s not done yet

  • Chapter-length rendering jobs are in progress; current renders run on CPU while the GPU path is being wired up.
  • Character-aware narration (different voices per speaker) is a later phase.
  • The public demo, when it ships, will play pre-rendered samples — a website can't run a GPU render farm, and the demo page will say exactly that.