Inside Voice

Inspiration

Reading is easy. Remembering it is hard.

Rereading feels productive, but retrieval—trying to explain what you just read from memory—is far better for long-term retention.

Most reading apps optimize for finishing faster. Study apps add recall, but outside the reading experience.

Inside Voice puts retrieval directly inside the reader.

What it does

• Synced narration: Highlights the exact word being spoken. Tap a word to jump audio there.
• Recall: After a few paragraphs, recall what happened before seeing a summary.
• Ask the book: AI answers using only text you have already read, preventing spoilers.
• Speed reading: RSVP mode from 100–1000 WPM.
• Notes: Save or summarize paragraphs.
• Customization: Fonts, spacing, themes, margins, voices, and more.
• Runs on iPhone, iPad, and Mac.
• Most features work offline with no account.

How we built it

The core is:

timeline: [(word, startTime, endTime)]

Everything follows one rule:

Current playback time determines the highlighted word.

This makes seeking, speed changes, pause/resume, and highlighting reliable.

Built with:

• Swift 6 + SwiftUI
• TextKit 2 pagination
• AVSpeechSynthesizer + AVAudioPlayer
• Kokoro-82M/CoreML neural voice
• EPUB parsing written in Swift
• AI restricted to already-read text
• 258 core tests

Challenges

• Creating accurate word-level audio timestamps
• Preventing the AI from spoiling unread text
• Making recall useful without constantly interrupting reading
• Handling Swift 6 concurrency with AVFoundation
• Keeping huge books fast through lazy loading

Accomplishments

• War and Peace opens in about 16ms
• Word highlighting stays synced across seeking and playback speeds
• Recall works with cloud AI, Apple's on-device model, or no AI
• 258 tests with zero dependencies in the reading core
• One app target across iPhone, iPad, and Mac

What we learned

A strong invariant simplifies everything.

For us:

highlight = function(currentTime)

We also learned that interaction order matters: recall first, reveal second.

What's next

• Test neural voices on real devices
• Background audio + lock-screen controls
• Disk caching
• Voice-based recall
• Bookmarks, highlights, and search
• App Store release

Built With

• apple-foundation-models
• avfoundation
• avspeechsynthesizer
• coreml
• epub
• fluidaudio
• google-gemini
• ios
• ipados
• keychain
• kokoro-82m
• macos
• swift
• swift-6
• swift-package-manager
• swift-testing
• swiftui
• textkit-2
• xcode
• xcodegen