Audiobook PDF Reader

A client-side app that turns any PDF into an audiobook—no server, no API key, no account.

Live: audiobook.vedgupta.in · Source: github.com/innovatorved/audiobook

What it does

• Upload PDFs and keep them in a local library
• Listen with an offline neural TTS voice (Kokoro via ONNX Runtime Web)
• Follow word-by-word highlighting synced to playback
• Click any sentence to seek to that point
• Resume where you left off

Everything runs in the browser. PDFs never leave your device.

Word-level sync

This was the most interesting part to build.

1. pdfjs-dist extracts text with character positions per page
2. sentence-splitter chunks the text into TTS-sized pieces
3. Each chunk is synthesized ahead of playback; the model returns audio + character-to-millisecond timing data
4. A requestAnimationFrame loop compares the current audio time against the timing map and updates the active word pointer
5. Clicking a word resolves its sentence, seeks the audio to that timestamp, and resumes instantly

Stack

React 19 · TypeScript · pdfjs-dist · kitten-tts-js (ONNX Runtime Web) · Tesseract.js · Zustand · Dexie (IndexedDB) · TanStack Virtual · Vite · Bun

Open audiobook.vedgupta.in , upload a PDF, and press play.

— Ved Gupta