Private, on-device audiobooks

Whisperbook

An offline-first Android app that turns local EPUB and PDF books into synchronized, multi-voice audiobooks without an account, cloud upload or runtime network access.

Back to selected work

Role: Chakib designed and built the Android product end to end: document import and extraction, chapter preparation, explainable speaker attribution, local voice casting, progressive Supertonic 3 synthesis, Media3 playback, synchronized reading and offline MP3 export.
Current status: Public v0.1 Android release for arm64 devices on Android 8.0 or newer. The app imports local EPUB/PDF files, narrates with bundled Supertonic 3 voices, supports English, French and Arabic book settings, keeps playback synchronized with passages, and exports one local MP3.

Whisperbook Android now-playing screen with offline status, chapter, speaker, playback, seek, timer and voice-cast controls — Now Playing keeps the active speaker, chapter controls and offline state visible on the production Compose surface.

Context and problem

Turning a book into a listenable audiobook usually means uploading private reading material, waiting for the whole book to render, and losing control over voices, chapters and corrections. Whisperbook keeps import, attribution, synthesis, playback and export on the device.

Ownership and contribution

Chakib designed and built the Android product end to end: document import and extraction, chapter preparation, explainable speaker attribution, local voice casting, progressive Supertonic 3 synthesis, Media3 playback, synchronized reading and offline MP3 export.

Responsibilities

Built the Kotlin/Compose product surfaces and Woodland Paper Theatre design system.
Implemented local EPUB/PDF extraction, bundled OCR fallback and durable per-book state.
Integrated Supertonic 3 through sherpa-onnx with local voice previews and per-book or per-chapter casts.
Built progressive chapter preparation, Media3 playback, read-along checkpoints and chapter continuation.
Added cancellable offline MP3 export that reuses finalized narration and generates only missing audio.

Architecture and data flow

A single-module Kotlin/Compose app copies the selected publication into private storage, extracts EPUB or PDF text, detects chapters and speakers, persists the library in Room, and prepares opening audio through WorkManager. Supertonic 3 generates finalized local WAV segments that Media3 can play progressively while later chapters continue preparing.

Local EPUB or PDF
Private extraction and chapter detection
Speaker attribution and voice cast
Progressive Supertonic 3 synthesis
Media3 playback and synchronized passages
Offline book-level MP3 export

Engineering decisions

Remove runtime networking entirely.

The Android manifest has no INTERNET or ACCESS_NETWORK_STATE permission, so imported books, generated audio and listening history stay inside the device boundary.Alternative not selected: Requiring a hosted model, account or cloud fallback for core narration.

Stream finalized local segments instead of waiting for the whole book.

The opening chapter becomes listenable first while later chapters are prepared sequentially, without sacrificing cache, seek or restart behavior.Alternative not selected: Blocking playback until every chapter has been synthesized.

Reliability, security and evaluation

Room, DataStore and WorkManager persist library, settings and preparation checkpoints across restarts.
Finalized WAV microsegments remain the common cache, playback and export representation.
JVM and connected Android tests cover import, synthesis, playback, navigation, voice selection and chapter continuation on the production Compose surfaces.

Outcome and status

Public v0.1 Android release for arm64 devices on Android 8.0 or newer. The app imports local EPUB/PDF files, narrates with bundled Supertonic 3 voices, supports English, French and Arabic book settings, keeps playback synchronized with passages, and exports one local MP3.

Technology groups

Android

Kotlin, Jetpack Compose, Room, DataStore, WorkManager

Audio

Supertonic 3, sherpa-onnx, Media3, AudioTrack, FFmpegKit

Documents

EPUB, PDF, ML Kit OCR, Storage Access Framework

Proof and source material

Whisperbook synchronized read-along screen with speaker-labelled passages and playback controls — Read Along follows the active passage and keeps manual speaker correction available.

Whisperbook voice-cast screen with book language and local character voice assignments — Voice Cast manages per-book language and previewable local voices without leaving the device.

Whisperbook offline library showing continue listening and additional local books — The Library restores each book's own listening position and preparation state.

Whisperbook source Latest Android release