Narrate — Read Aloud
Upload a manuscript and have it read aloud right in your browser — a proofreading tool that also runs a genuinely natural neural voice, no server or paid API involved.
What it is
A tool I built for myself first: drop in a manuscript, have it read back to you. Listening to your own writing catches things your eyes tend to skim past — awkward rhythm, repeated words, sentences that don't quite say what you meant. It lives right here on this site at /projects/narrate.
How it works
Everything happens right in your browser — nothing is ever uploaded anywhere:
- Reading your file — text, Markdown, and Word documents are all read directly on your device.
- Two ways to listen:
- Your device's voice — reads each paragraph back instantly, using whatever voice is already built into your computer or phone.
- A better voice, if you want it — an optional upgrade to a small, efficient AI voice model that runs entirely on your own device. No server, no ongoing cost, and no waiting on a distant data center. It downloads once, is remembered afterward, and shows its progress honestly along the way — falling back gracefully to your device's own voice if the download doesn't finish.
- Both options share the same play, pause, and skip-by-paragraph controls, which is what makes this genuinely useful for proofreading instead of just playback.
The path here
This started life as a more complicated, server-based version. Getting a server reliably up and running turned into its own project before the tool itself could even be tried — so I rebuilt it to run entirely on-device instead, with no server to set up or maintain at all. That change also happened to be the more efficient, sustainable way to build it: a small, well-chosen AI voice model beat needing a server running around the clock.
Formats
Text files, Markdown, and Word documents work today. PDF support is a natural next step. A couple of other document formats are intentionally left out for now, since they convert cleanly into formats already supported.
What's next
PDF support, and possibly an optional "save as an audio file" feature down the line, for anyone who'd rather have a downloadable file than listen right in the browser.