Privacy Policy

What we collect

• Account: your e-mail address, your name and company details if you give them, and your password (hashed, held by our authentication provider). With Google sign-in, the name and e-mail Google shares.
• API keys: only a SHA-256 hash of the key and its first 16 characters. We never hold the key itself; it is shown once when created.
• Usage records: time, endpoint, model, audio seconds or character count, processing time and the key used. No text, no audio.
• What you save: your saved voices' names, reference clips and transcripts with the time of consent; the descriptions of voices you designed; the terms in your pronunciation dictionary.
• Site statistics: the page viewed, language, country and the referring site's name. We count visitors by a daily-salted hash of IP address and browser; the IP address is never stored.
• Technical logs: IP address, browser, request path and status code. Kept for up to 90 days.

Text, audio and generated speech

Text you send to Alania is processed in memory and the speech is returned to you; we store neither the text nor the generated audio. Audio you send to Duyu is processed in memory, the transcript is returned and the audio is discarded right away; we do not store the transcript.

Live streaming works the same way: chunks are processed, never stored. Only seconds or characters are recorded.

For quality, we may check on our own server, with Duyu, that the generated speech matches the text; the result of that check is not stored either.

Your recent work in the Studio (speech you generated and transcripts you made) is kept only in your browser and never sent to our server.

Style instructions and voice descriptions

Alania-2 follows instructions in English. If you write a style instruction or voice description that is not in English, or a long one, only that instruction (never the text to be spoken) is sent to an AI service provider (EU) to be rewritten as a short English tag, or to a translation service if that does not answer. Short English instructions are never sent anywhere.

Voice cloning and biometric data

When you save a voice, its reference clip is kept in private storage only your account can reach, until you delete it.

For a professional clone, your uploads and live reading are used to train your voice. We also compare voice prints to confirm that the reader is the same speaker as in the uploads. We ask for, and record, your explicit consent to this biometric comparison separately. Voice prints are computed in memory and never stored; only the similarity score is kept.

Professional clone recordings are deleted 7 days after training ends, within 24 hours for an application left unfinished, and at once if verification fails or you delete the application. The trained voice model is deleted when you delete the voice.

Where data is processed

The models are ours and currently run on our own servers hosted in the Netherlands (EU). Professional clone training runs on our server in Sweden (EU). Account, usage and saved-voice data are stored in Germany (EU). The website is served through a global content delivery network; its server-side functions run in Germany.

Enterprise customers can license the models and run them on their own servers or cloud accounts; text and audio then never reach our servers.

Model training

The text and audio you send are not used to train our general models. If we ever want to change that, we will announce it here first and offer an opt-out. Enterprise agreements guarantee this in writing.

Cookies

Session cookies for signing in, a language cookie and a theme preference kept in the browser. Site statistics use no cookies. No advertising or third-party analytics cookies.

Service providers

• Cloud infrastructure providers: model servers (Netherlands, EU) and professional clone training (Sweden, EU).
• A managed database, file storage and authentication provider: account data, saved voices and verification e-mails (EU).
• An AI and translation service provider: rewriting style instructions and voice descriptions that are not in English or are long (EU; the fallback translation service may process outside the EU).
• A web hosting provider: the website.

Retention

Account and usage data, saved voices and your pronunciation dictionary are kept while the account exists and deleted within 30 days after it is closed. You can delete a saved voice or a dictionary term yourself at any time. Technical logs are kept for up to 90 days.

Your rights

Under KVKK and GDPR you can access, correct, delete and object to the processing of your data, and withdraw your consent to the biometric comparison. To close your account or make a request, reach us at patientdesk.ai.

Security

All traffic is encrypted with TLS. Keys are stored hashed. Saved voices and clone recordings sit in private storage only their owner can reach. Access to the servers is limited to authorised staff.

Apps

The privacy notes for the Eshlik desktop app and the Antalia Android engine are on their own pages.

Changes

If we update this policy we publish it here with its date. For the terms of use see the Terms of Service.