Where does your audio go? Questions to ask before you choose an AI dubbing service

Published 8 October 2026

When you upload a video to an AI dubbing service, your voice, your guests' voices and your unreleased content go somewhere; before you sign, find out exactly where and what happens to them.

Why does it matter where your audio goes?

A dubbing job is not just a file conversion. The service hears every word in your video, analyses the voices of everyone who speaks, and often builds a voice profile that can say new things in those voices. That is sensitive material.

It can include unreleased episodes, client work under NDA, interviews with people who agreed to appear only on your channel, and voices that may count as biometric data under data protection law. If any of this leaks or is reused, the problem is yours as well as the vendor's. Under GDPR and KVKK you are usually the data controller, which means you are responsible for choosing a processor that protects the data properly.

Shared cloud or dedicated server: what is the difference?

Most AI dubbing services run on shared infrastructure. Files from many customers are processed on the same pool of machines. Some steps, such as transcription, translation or voice synthesis, may be sent to external AI providers through an API. Each of those providers is another place your audio travels to, with its own storage and retention rules.

A dedicated setup works differently. Your jobs run on a server assigned only to you. Your audio and video do not leave that server, and no other customer's work runs on it. This makes the answers to most of the questions below simpler, because there are fewer parties involved.

Neither model is automatically safe or unsafe. A shared service can be well run, and a dedicated one can be careless. What matters is that the vendor can answer clearly and put the answers in writing.

What should you ask an AI dubbing vendor?

Use this checklist when you compare services. A good vendor will answer each point without hesitation.

1. Where is my audio processed and stored? Which country, which provider, and is it shared hardware or a server dedicated to you?
2. Who are your sub-processors? Ask for the list of every third party that touches your data, including external AI services, cloud hosts and analytics or telemetry tools.
3. Is my content used to train models? Ask whether uploads, transcripts or voice profiles are used to train or improve any model, yours or a third party's. If yes, is there an opt-out, and is it in the contract?
4. How long do you keep my data, and how is it deleted? Ask about source files, intermediate files, finished dubs and backups separately.
5. Who owns the voice profile? Ask whether a profile built from your speakers can be used for other customers, for demos or for any other purpose.
6. What happens when the contract ends? Is everything deleted, including voice references and profiles? Will you get written confirmation?
7. Is there a data processing agreement? Under GDPR Article 28, a processor working for you needs a written contract that covers these points. Ask to see it before you sign.
8. How do I report a security issue? Look for a published security contact, such as a security.txt file, and ask how quickly reported issues are handled.

Two more questions are worth adding: how do you handle consent for the speakers in my videos, and which licences cover the AI models you use? A vendor that runs models without a licence for commercial use can create problems for you later.

What about labelling AI-generated content?

The EU Artificial Intelligence Act (Regulation on AI) includes transparency rules in Article 50. In broad terms, providers of AI systems that generate synthetic audio, image or video must make sure outputs are marked as artificially generated in a machine-readable way. Those who publish deepfake-style content, meaning content that resembles real people and could be taken as authentic, must disclose that it was generated or manipulated by AI.

How these rules apply to a particular dubbed video depends on the details, and guidance on implementation is still developing. Two practical steps make sense regardless. First, ask your vendor how its outputs are marked. Second, tell your audience when a video is dubbed with AI, for example in the description or on screen. Viewers tend to accept AI dubbing more easily when they are told about it.

This section is general information, not legal advice.

How does Dublayer answer these questions?

Dublayer runs each customer's work on a dedicated GPU server, and your audio and video do not leave that server. We do not use external AI services or telemetry. Your uploads are never used to train any model. Voice profiles are used only for your videos, are never shared or sold, and are deleted when your plan ends. You are the data controller, we are the data processor, and the data processing agreement is part of the contract. When the contract ends, project data, voice references and profiles are deleted and you receive a deletion record. We use only open models licensed for commercial use, and outputs are marked as AI-generated in line with Article 50 of the AI Act. Security issues can be reported via /.well-known/security.txt or hello@dublayer.studio; our security page and acceptable use policy have more detail. To talk about your own videos, request a quote and we'll set up a meeting.