Privacy policy

Account: name, email, password hash. Content: the text and recordings you upload, and the audio and transcripts produced. Usage: which features you use and how much, signed-in devices and IP address.

Signals: which takes you keep, regenerate, star or download, and the edits you make to transcripts and translations.

Free access comes with one condition: we use what you put into Voxy, and what it produces, to train, test and improve our speech recognition, speech synthesis, translation and voice models. That's how Voxy gets better at Mongolian for everyone.

That includes text you type, audio you upload or record, generated audio, transcripts and, most valuably, your corrections — every word you fix teaches the model something.

• We strip your name, email and account details before anything enters a training set.
• We filter out obvious personal details such as phone, registration and card numbers where we can.
• Training runs on our own servers. We don't sell your content or give it to other companies to train their models.
• A small number of our staff may listen to or read samples to check quality, under strict confidentiality.
• Your cloned voice stays yours: training may improve voices in general, but nobody else can generate speech in your voice unless you publish it.

Delete a generation, transcript, voice or your whole account, and we delete its training copy too and leave it out of future training. We can't remove what an already-trained model has learned.

Please don't upload anything you wouldn't want used this way, such as health records, passwords, or private conversations recorded without people's consent.