content · 2026
ESL Story Engine
An automated ESL podcast studio — Claude writes character-driven stories, Kokoro voices them, FFmpeg edits, YouTube ships them daily.
Three characters — a British narrator, a chatty New Yorker, and a warm London host — walk English learners through a new story every day. The whole thing runs unattended on a single workstation: Claude writes the script and weaves in vocabulary for the level, Kokoro voices the cast locally at zero per-minute cost, WhisperX aligns the audio, and Pillow + FFmpeg stitch the video. A Gemini-generated thumbnail lands last, then the pipeline schedules the episode straight to YouTube. Zero SaaS in the loop. Zero missed publish days.
- Three-character AI cast — Jake, Oliver, Sophie — each with a fixed voice through Kokoro's local TTS, no per-minute billing
- Long-form podcast episode plus a same-day Short cut from the same story, both scheduled to YouTube in one run
- Vocabulary layer weaves target words (health, workplace, travel, small-talk) into the script so learners get real language in real context
- Hardened for the ugly parts — atomic state writes, lock-file concurrency, slot rollback on publisher failure, Windows console encoding fixes
- Windows Task Scheduler drives the cron; the workstation ships content while the operator sleeps
- Claude API
- Kokoro TTS
- WhisperX
- FFmpeg
- Gemini
- YouTube Data API
The three-character cast has never called in sick, asked for a raise, or missed a Monday morning.
