AI Transparency Notice

What Voxify Studio generates with artificial intelligence, how that output is marked, and how anyone can check it.

Effective date: 1 August 2026 · Last updated: 4 August 2026 · Version 1.1

This notice is published under Article 50, including Article 50(2), of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689), whose transparency obligations apply from 2 August 2026. It explains which parts of the audio produced by Voxify Studio are generated by AI, what machine-readable marking we apply to that audio, and what you — as the person publishing the audio — still need to do.

1. What is generated by AI

Voxify Studio is a platform for producing audio advertisements. A finished advertisement can combine several elements, and not all of them are AI-generated:

Voice-over — AI-generated. All spoken audio is synthetic speech produced by text-to-speech models. No human recording is involved unless you upload one yourself.
Ad script — AI-assisted. Scripts can be drafted or rewritten by large language models from your brief, product page or prompt. You remain the author and are responsible for the final wording.
Sound effects — AI-generated when generated on request. Sound effects created from a text description are AI-generated. Sound effects picked from a pre-existing library are not.
Music — usually not AI-generated. Music beds are, as a rule, licensed human-composed recordings from our music partner’s catalogue. AI-generated music is available only where it has been expressly enabled for your account.
Audio you upload — not AI-generated. Voice recordings, music or effects you upload yourself are your material. We do not mark them as AI-generated.
Mixing and mastering — automated processing, not generation. Levels, ducking, loudness normalisation and mastering are signal processing applied to the elements above.

A typical finished advertisement is therefore a composite: an AI-generated voice over a human-composed music bed. We mark it accordingly, and we do not claim that material is AI-generated when it is not.

2. How we mark AI-generated audio

Voxify uses two complementary machine-readable marking layers. Every audio file the platform generates carries embedded metadata identifying it as artificially generated. In addition, each final rendered audio variant carries an imperceptible audio watermark embedded in the final mix before encoding.

The metadata layer applies to finished advertisements, individual voice-over clips, generated sound effects, generated music and voice previews. It uses the IPTC “Digital Source Type” vocabulary — the same controlled vocabulary used by the C2PA / Content Credentials standard — so that it can be read by any tool that understands the standard, not only by our own software:

trainedAlgorithmicMedia — for material created entirely by a generative model (a voice-over clip, a generated sound effect, AI-generated music).
compositeSynthetic — for a finished advertisement that mixes AI-generated material with other elements, such as a licensed music bed.

Alongside it we record, in the same file: that the content is AI-generated, that Voxify Studio produced it, the marking version, the type of content, and a link back to this notice. The marking is also written into standard fields, so it is visible in ordinary media players and tools such as ffprobe — not only to specialist software.

The watermark is a separate signal-level layer. Its purpose is to keep a Voxify-origin indicator in the audio even when a downstream workflow strips file metadata. It does not contain all of the richer descriptors held in metadata; the two layers are checked and reported independently.

2.1 How to check a file

Any recipient can inspect the metadata of a file produced by Voxify Studio with standard tooling, for example:

ffprobe -show_entries format_tags yourfile.mp3

A marked file reports IPTC_DIGITALSOURCETYPE, AI_GENERATED=true and AI_GENERATOR=Voxify Studio. You can also use the public Voxify AI Audio Detector to check the watermark and metadata layers. If you need to verify a file and cannot do so, contact us at support@voxify.studio.

3. Limitations you should know about

We state these plainly rather than overstate what the marking achieves:

Metadata can be removed. Re-encoding, editing or uploading audio to a third-party platform may strip embedded metadata. A file that has lost its marking is not thereby a file that was not AI-generated.
Watermark detection is not universal or conclusive. The detector checks for the Voxify watermark, not marks from every AI provider. Severe transformation may affect detection, and a negative result is not proof that the audio is human-created.
Marking is not a cryptographic guarantee. The current marking is not digitally signed, so it proves origin only to the extent the file has not been tampered with. We intend to add signed provenance manifests as the supporting certification ecosystem becomes available to us.
Upstream providers may apply their own marking. Some of the AI models we use may add their own watermarks to their output. Where they do, that marking is theirs, is applied under their control, and may not survive our mixing and mastering. Our own marking does not depend on it.
Marking describes the file, not the claims in it. It says how the audio was produced. It says nothing about whether the advertising message is accurate or lawful.

4. Your obligations when you publish

Voxify Studio, as the provider of the AI system, is responsible for marking the output. You, as the person who publishes or broadcasts the audio, may have your own separate disclosure duties — in particular under Article 50(4) of the AI Act, and under advertising, broadcasting and consumer-protection rules in each country where the advertisement runs.

In practice this means you should:

keep the embedded marking intact where your workflow allows it;
make any audible or written disclosure that the applicable rules or the platform you publish on require, in a clear and distinguishable way, at the latest on first exposure to the listener;
never present synthetic speech as a recording of a real, identified person without that person’s documented consent — see our Acceptable Use Policy.

We provide the marking and the means to verify it. We cannot make the disclosure on your behalf, because only you control where and how the advertisement is published.

5. AI models we use

Voxify Studio builds on third-party AI models, including text-to-speech, sound-effect and language models from established providers. We select providers whose terms permit commercial use of the output for advertising, and we pass through the resulting rights to you under our Terms of Service. The list of providers may change as the Service evolves; current details are available on request at support@voxify.studio.

6. Changes to this notice

We will update this notice as our marking technology and the applicable standards develop — including when we add signed provenance manifests or materially change the watermarking system. Material changes will be reflected in the version and date at the top of this page.

7. Contact

Questions about AI transparency, or a request to verify a specific file: support@voxify.studio.
Voxify Studio is operated by GZK CONSULTING S.R.L., Romania.