This policy describes what Orea Studio collects, where it goes, and what you can ask us to do about it. It is written to match what the software actually does. If something here does not match your experience, tell us: that is a bug in one or the other.
Galvan Group B.V. operates Orea Studio as part of the Galvan Intelligence platform, and is the data controller for the personal data described here. Reach us at [email protected] for any question or request in this policy.
| Data | Why | Kept |
|---|---|---|
| Email address, name, organisation, what you plan to transcribe | Your account, the sign-in link, and deciding plan requests | While your account exists |
| The audio you upload, the 16 kHz copy we make of it, and the transcript with speaker labels and timings | Producing your transcript, and possibly improving Orea (below) | Until you delete the recording, and in the training corpus after that (below) |
| Every edit you make to a transcript or translation (the text before and after) | Your editing history, and improving Orea | Same as the recording |
| Minutes of audio per month, job timings, model version used | Your plan's allowance and our capacity | Indefinitely, as counts |
| Sign-in and last-seen times | Support and abuse handling | While your account exists |
| What you do in the app, linked to your account number: pages you open, uploads started and finished, transcripts opened, exports, shares and re-runs, with the file size and type, recording length and the export format. Never the audio or the transcript text | Seeing how Studio is used and where people get stuck, so we can fix it | Up to 5 years, and removed with your account |
A voice is personal data. A recording identifies the people speaking in it whatever the transcript says, so audio cannot be anonymised by removing names. Only upload recordings you are entitled to process: your own, or ones where the people recorded know and agree. That is your responsibility as the person uploading; ours starts once the file is here.
Orea is a speech model for Papiamentu that gets better with real speech and real corrections. Any recording you upload, the transcript it produces, and any correction you make may be used to train and evaluate Orea and related Galvan Intelligence models. There is no per-upload opt-out; it is the deal the Free plan is built on and it is stated on the upload screen. Not every recording ends up in training; which ones do is a curation decision on our side.
What that means in practice: a copy of the audio and the transcript pair can be kept in a training corpus after you delete the recording from your account, stored on Galvan's own infrastructure in the EU, accessible only to Galvan Group staff working on the model. It is not sold, not shared with other customers, and not used to identify or profile the people speaking. Contact us if a specific recording must be excluded after the fact and we will remove it from the corpus where technically possible; a model already trained on it cannot be untrained.
Your file is stored on a Galvan server in the EU (Hetzner, Germany). For transcription it is sent to a GPU server Galvan rents (RunPod, EU data centre) that runs Orea and the speaker separation model; the audio is deleted from that server as soon as the job is done. If you choose a translation, the transcript text (not the audio) is sent to Galvan's own Chuchubi model running on Hugging Face Inference Endpoints in the EU. No third party receives your audio to train their own models.
Identity is handled by Galvan's own Keycloak server; the sign-in link is sent through Mailgun (EU region). One account works here and on chat.galvangroup.co. We send email about your account, a finished or failed transcription, and plan changes. Nothing promotional.
A share link makes a read-only copy of the transcript readable by anyone holding the link, without an account. Revoking it in the editor stops it at once. The audio is never shared.
You can delete any recording from the app, which removes the audio, the transcript and the edits from your account. You can ask for a copy of your data, correction, deletion of your account, or restriction of processing at [email protected]. Deletion of an account removes everything above except the training-corpus copy described under Training, which you can also ask about. You can complain to the Dutch Autoriteit Persoonsgegevens or the Curaçao supervisory authority.
One session cookie for sign-in. No advertising, no cross-site tracking. The usage events above are counted by Aptabase, which Galvan runs itself on its own server in the EU (Hetzner, Germany); no outside analytics company receives them. A random id for the open browser tab is kept in session storage and disappears when you close the tab.