Most of it stays on your device
Your settings, the pages and PDFs you open, scanned pages being recognised, and the language of what you read are all handled in your own browser.
Your settings, the pages and PDFs you open, scanned pages being recognised, and the language of what you read are all handled in your own browser.
To turn text into speech, the sentences you ask to hear go to Narractor’s voice server, with your chosen voice, the language and a short-lived login token. Audio comes back. That’s all that leaves.
The voice poll sets a random token so each person votes once. It holds no name, no email and no readable address.
No analytics, no ad trackers, no browsing history, no selling of anything to anyone. An account is optional, never required.
| What | Where it’s kept | For how long |
|---|---|---|
| Your settings: voice, speed, volume, reader and theme choices | Your browser, on your device | Until you change them or remove Narractor |
| The text of a page or selection you ask to hear | In the page, while it’s read | Until you stop or leave the page |
| The sentences you listen to, your voice and the language | Sent through narractor.com to the voice server | Kept in the voice server’s memory until it stops, which it does when nobody has listened for a while |
| Audio of sentences you’ve heard | In the player’s memory | Until you stop reading |
| Your place while listening: the page, its sentences, the position | Your browser’s session storage | Until you stop, or the browser closes |
| PDFs you open, and scanned pages recognised | In the PDF reader tab | Until you close it |
| The language of what you read | Worked out by your browser, on your device | Not kept |
| Your vote in the voice poll, with salted fingerprints | This website’s server | With the poll’s results |
| A login for this browser: the free trial or your account | Narractor’s own storage in your browser; narractor.com keeps only scrambled forms | Until you log out or remove Narractor |
| How many seconds of audio were made for a login | narractor.com | A trial’s total while it lasts; an account’s daily counts for 30 days |
| Your email address, if you make an account, and your Google account ID if you log in with Google | narractor.com | Until you delete the account |
| For each browser you log in or connect Narractor in: its name and system (like “Chrome on macOS”), when it logged in and was last used, and your display name if you set one | narractor.com | Until you log it out, disconnect it or delete the account |
| Account activity: sign-ups, logins, connections and security events, without what you read | narractor.com | With the account; an admin action on a deleted account keeps only its email in the log |
The narractor_session cookie, when you log in on this website | Your browser (not readable by scripts) | 30 days, or until you log out |
The narractor_voter cookie | Your browser | One year |
| Contact and voice-request forms | Your clipboard, when you copy them | Not sent to us |
Narractor’s script is present on the pages you visit so that its selection menu and keyboard shortcuts work, but it only reads a page’s text when you ask it to: with the N button, a shortcut, the right-click menu or the selection menu. It reads the article (or what you selected), not other tabs, and keeps no history of what you’ve read.
Before reading, Narractor works out the text’s language with your browser’s own language detection, which runs on your device. Highlighting the sentence being read never changes the page itself.
When you open a PDF in Narractor, the extension downloads it itself, sending your browser’s cookies for that site where your browser allows it, so it can open PDFs you’re signed in to see. It’s taken apart and rebuilt into an article inside your browser. Figures are cut out of the pages there as well.
Scanned pages are read with text recognition that runs on your computer; the pictures of the pages never leave it. As with any page, only the sentences you listen to go to the voice server. A PDF you pick from your own files is never uploaded anywhere.
Speech is made by Narractor’s voice server. For each sentence you listen to, it receives the sentence’s text, the voice you chose and the language, and sends back the audio. It keeps the most recent few hundred finished sentences in memory, so jumping back or replaying is instant, until it restarts. narractor.com also counts, for running the service, how much audio was made and how long that took. Neither logs the text.
The first time you press play, Narractor asks how you want to listen. Without an account you can try it free for 30 minutes in total; with a free account, three hours a day across every browser you connect. Either way your browser gets a login from narractor.com, kept in Narractor’s own storage where web pages can’t reach it, and each sentence goes to narractor.com with a short-lived token that says only “a trial” or “an account” and which one. narractor.com checks it, passes the sentence on to the voice server and counts the seconds of audio made for it, never what was read.
An account is just your email address. You log in on narractor.com with Google or with a 6-digit code sent to your email, then connect Narractor with one click. With Google, Google tells us only your email address, that it’s verified, and an ID for your Google account, so the two stay linked; we never see your Google password, name or contacts. There’s no Narractor password either; there’s no password, and the email is used only to log you in, never for marketing. On your account page you can set a display name, see every browser that’s logged in or connected and log any of them out, or delete the account, which removes your email immediately.
To run the service and stop abuse, Narractor’s administrators can see account emails and display names, the browsers each account uses, login activity and how much each account listens. They can log an account out or suspend it. They never see what anyone reads, and every admin action is recorded.
The voice server runs on graphics cards rented from RunPod, in data centres in the EU, and only starts when someone is listening. Your browser never talks to it directly: each sentence goes over an encrypted connection to narractor.com, which checks the login, adds Narractor’s own key and passes it on, and the audio comes back the same way. RunPod runs those machines, so the sentences being spoken pass through its systems, as they would at any host; Narractor asks it to keep nothing. This website and the accounts run on Narractor’s own server at Hetzner, in Germany.
Your settings are saved in your browser’s storage for extensions, on your device. While you listen, your place (the page’s title and address, its sentences and where you are) is kept in your browser’s session storage, so reading can carry on if the browser pauses the player; it’s cleared when you stop, or when the browser closes. Removing Narractor removes all of it.
The poll on Request a voice is the only thing on this website that keeps anything. When you vote, it stores your choice together with three salted, one-way fingerprints, used only so each person votes once in each poll:
narractor_voter cookieYour IP address itself is never stored, and fingerprints can’t be turned back into one. The poll doesn’t ask for your name or email.
The website has no analytics and no advertising trackers, and sets no cookies other than the poll’s and, if you log in, your login session’s. Its fonts, narrator samples and pictures are served with the site itself, so visiting doesn’t contact other companies. The Stats page shows sample figures, not measurements of anyone’s listening.
The contact and voice-request forms aren’t connected to anything yet: “Copy” puts your message on your clipboard, and nothing is sent or saved. The article demo on the home page uses your browser’s own speech, whose voices come from your browser or system; some of those can use an online service, depending on the voice you pick.
The site is hosted through ChatGPT Sites. Like any web host, it receives technical details of each visit, such as your IP address and browser, to serve and protect the site, under its own policies. Links to Dudji.com and Wikimedia Commons lead to sites with their own practices.
This page changes when Narractor does, and before any change in where your information goes, such as the voice server moving online. The date at the top says when it last changed.
Questions? Prepare one on the Contact page (it copies your message for now), or find the maker at Dudji.com ↗.