manuscriptreader.com — Free, private text-to-speech for writers.
Paste a chapter, pick a voice, press play. Your text is read aloud in a natural AI voice, with sentence highlighting so you can follow along. Everything runs locally in your browser. No servers, no accounts, no tracking.
The site uses Kokoro, an open-source text-to-speech model with 82 million parameters. On first visit, the model (~86 MB) downloads and caches in your browser. After that, all speech generation happens on-device — your text is never sent anywhere.
- 14 natural AI voices (American and British, male and female)
- Sentence-by-sentence highlighting with auto-scroll
- Speed control
- Adjustable pauses between sentences and paragraphs
- Keyboard shortcuts (Space to play/pause, Esc to stop)
This is the core promise: your manuscript never leaves your browser.
- No servers or APIs process your text
- No analytics, cookies, or tracking of any kind
- No accounts or logins
- Nothing is saved — close the tab and it's gone
- The voice model is pre-trained and read-only; your text is not used for AI training
The entire site is a single HTML file. No build step, no dependencies, no backend.
- Kokoro TTS via kokoro-js (Apache 2.0)
- ONNX model runs client-side via WebAssembly
- Hosted on GitHub Pages
MIT