Privacy Policy

The short version

• Your file is never uploaded. The video or audio you choose is read and transcribed inside your own browser. This site has no server that could receive it.
• No cookies, no account, no profile. Nothing is stored in your browser about you — only three small entries that make the page work: the speech model it caches, which model you used last, and a marker that is deleted as soon as a transcription finishes. The section below says exactly what each one is.
• Page views are counted without cookies by Cloudflare Web Analytics, which is built not to identify individuals.
• No advertising is served on this site today. Third-party advertising is planned. Before any ad appears, this policy will be updated to describe it, and the date at the top of this page will change.

Your video and audio files

When you pick a file, the browser reads it directly from your device. The audio is decoded and the speech is transcribed in your browser, using a model that is also running in your browser. The file itself is never transmitted anywhere — there is no upload step, no queue, and no copy of it on any server belonging to this site.

You do not have to take that on trust. The whole application is a set of static files you can inspect, and the network activity of a transcription run is small and observable: apart from the model download described in the next section, the only request the page makes while transcribing is a short report to the site's own analytics endpoint. It contains no part of your file.

The one thing that is downloaded: the speech model

The speech recognition model does have to be fetched, because it is far too large to ship inside a web page. On the first transcription, your browser downloads the model weights from Hugging Face (huggingface.co, with the large files served from its CDN at us.aws.cdn.hf.co). This is an ordinary download of a public file — the same file every visitor receives. It carries nothing about you or your file.

Your browser then keeps that model in its own cache, so later transcriptions do not download it again. You can clear it at any time through your browser's site-data settings; the only consequence is that the next run downloads the model again.

Analytics

This site uses Cloudflare Web Analytics to count page views and measure how quickly pages load. The measurement script is served from static.cloudflareinsights.com and reports to this site's own address. It is included because it is the least intrusive option available: Cloudflare states that Web Analytics "does not use any client-side state, such as cookies or localStorage, to collect usage metrics" and that it does "not 'fingerprint' individuals via their IP address, User Agent string, or any other data for the purpose of displaying analytics" (Cloudflare, Web Analytics product documentation). What it reports is aggregate: page views, referrers, approximate country, and page-load timing. Cloudflare states that it does not log query strings, and that unsampled beacon data is retained for 7 days before being aggregated down.

There is no other analytics on this site — no Google Analytics, no advertising pixels, no session recording.

Cookies and storage in your browser

This site sets no cookies, it writes nothing to sessionStorage, and it creates no database in your browser. It writes three entries to localStorage. None of them is tied to your identity, none is sent anywhere, and none is used to recognise you on any other site.

• The cached model described above — a plain file cache, so you do not have to download the speech model twice.
• One preference, stored under the name reeltotext.model, recording which speech model you used last. It exists so the page opens on the model you used last time, and so a device that cannot fit the default model is not made to download it and fail on it again on every visit. Its value is that model's name, plus a short note when the page itself stepped down to a smaller model because your device ran out of memory for the one you asked for.
• A run marker, stored under the name reeltotext.run, written when a transcription starts and deleted the moment that run ends — whether it succeeded, failed, or was stopped by the page itself. It covers the one case where a run ends without the page being told: a tab the browser closes to free memory, a phone that restarts, a force-quit. If the marker is still there when you come back, the page tells you that the last run did not finish and removes it at that point, so you are never left with an unexplained failure.

Because there are no cookies and no tracking, there is nothing to opt out of on this site today, and no cookie banner is shown — there is nothing for one to ask about.

Advertising (planned, not yet served)

This site is free and has no accounts or paid tier. To cover the cost of the domain and hosting, third-party advertising is planned — most likely Google AdSense or a comparable network. No advertising is served on this site at the moment you are reading this. This section is here so the plan is not a surprise later.

When advertising does start, this is what will change, and this policy will be rewritten to say so plainly before it happens:

• Ad network code will load on the page, and those third-party vendors may set their own cookies or read device identifiers in order to select and measure ads. Those cookies are set by the vendor, not by this site.
• Google and other third-party vendors may use cookies to serve ads based on a visitor's prior visits to this site or to other sites.
• Where required by law, a consent notice will be shown before any non-essential cookie is set, and advertising cookies will not be set unless consent is given.
• You will be able to opt out of personalised advertising through Google's Ads Settings at adssettings.google.com, and you can read how Google uses data from sites that use its services at policies.google.com/technologies/partner-sites.
• The tools themselves will not change: your file will still be transcribed inside your browser and will still never be uploaded. Advertising does not need your file and will not receive it.

Until that happens, none of the above is in effect, and this page will not claim otherwise.

Other companies involved

Two service providers are involved in serving this site, and neither receives your file:

• Cloudflare hosts the site and provides the analytics described above. As with any web host, its servers process the requests needed to deliver pages to you, which necessarily includes your IP address.
• Hugging Face serves the speech model file described above. That is a download from its servers to your browser; nothing is sent back.

This site is independent and is not affiliated with TikTok, Instagram or Meta, nor with OpenAI, Hugging Face or the Transformers.js project.

Children

This site is a general-purpose utility and is not directed at children. It does not knowingly collect personal information from anyone, including children, because it does not collect personal information at all.

Your choices

Because nothing is collected and nothing is stored about you, there is no personal data held by this site to access, correct or delete. Clearing this site's data in your browser settings removes the cached model and the stored preference together; the next run will ask your browser for the model again. The run marker needs no action — it deletes itself as soon as a run ends. If you believe this site has handled something incorrectly, write to the address on the contact page and it will be looked at.

Changes to this policy

This policy will be updated whenever what the site actually does changes — in particular when advertising begins. The date at the top of this page is the date of the current version. The two tool pages used to state that the site ran no analytics; that stopped being true once Cloudflare Web Analytics was enabled, so those pages have been corrected alongside this one. This page itself used to state that the site wrote nothing to localStorage; that stopped being true when the page began remembering which speech model you used, so the storage section above was corrected at the same time. It has been corrected once more since: the page also records that a transcription has started, so that a run the browser ended without telling the page can be explained on the next visit.

Contact

Questions about this policy: see the contact page.