Ahmad ZushanAI · Systems · Platform

Home/2026 builds/№ 38

38 · Voice & language

Folio — any PDF as a readable, talking book

A Node service with Python workers: PyMuPDF and OpenCV ingest pages, PaddleOCR reads scans, a layout engine turns lines into a book and the browser paginates it by measuring real text. Narration and word meanings run locally with Piper and a 2-billion-parameter Qwen model, so voicing a whole book costs nothing.

PeriodSep 2026StatusPaused · server moveAreaVoice & languageDomainbookreader.zooshan.com

Highlights

  1. A scan and its digital twin produce byte-identical book structure after OCR.
  2. Piper voices run at eleven times real time on CPU; sentence-at-a-time synthesis gives karaoke-style highlighting.
  3. Built on a live-voice skill where the slide deck became a book: interrupt the reading and ask about the passage.
  4. Behind basic auth on purpose, because the upload endpoint is open and OCR eats four cores.