Skip to content
Heldenflug
Back to my audio stories

Voice consent

# Consent to process a voice and to create a voice clone

Provider: Tiamat UG (haftungsbeschränkt), An der Strusbek 12, 22926 Ahrensburg, Germany. Contact: [hallo@heldenflug.de](mailto:hallo@heldenflug.de). The service is offered as Heldenflug at [heldenflug.de](https://heldenflug.de).

Version: `hoerspiel/einwilligung-stimme.md@v2`

## What you are asked to allow

You can release a voice for your family's personal audio stories. You record three short lines: everyday, excited and gentle. The three recordings form one voice profile with three expressions. A revocation applies to the whole profile, not only to one expression.

The recordings are analysed technically. From those features a voice clone is created — a digital model of this voice.

The clone can then speak new lines. The person whose voice is used may never have said those lines. The clone is not a live recording of those lines, and the person is not speaking them at the moment of playback. We say this in plain language before recording (Art. 50 AI Act).

The clone is used only to produce episodes for your Heldenflug audio-story subscription. It must not be used for advertising, political statements, deception, identity checks, speaker profiling or other products.

## What data is processed

We process:

  • the uploaded speech sample including its audio data,
  • technical features of the voice and the model identifier created from them,
  • the label you enter for the voice, for example “Mum” or “Grandad”,
  • your consent, the time, the text version and whether you may act for another person,
  • the finished audio tracks in which the clone is used for an episode.

A voice can make a person recognisable. If the voice is processed to uniquely identify someone or to build an individual voice profile, this is biometric data. Biometric data are specially protected under Art. 9 GDPR. This consent is therefore worded expressly.

## Where processing takes place

The speech sample is transmitted encrypted to Heldenflug infrastructure and stored on servers of netcup GmbH in Germany. To create the clone it is sent over an encrypted connection to the ComfyUI router at `https://image-ai-router.moravene.com`. The router runs the versioned cloning process internally approved for new voice profiles. An existing voice profile stays pinned to the process approved when it was created.

The router is operated by Tiamat UG (haftungsbeschränkt) itself. Tiamat remains responsible for processing the voice. The GPU compute for speech synthesis runs in containers of the US provider Modal (A10G); Modal processes the data as Tiamat’s processor. Communication with the router is authenticated server-side with the `X-API-Key` header.

Because Modal is based in the USA, a third-country transfer may occur. There is a processor agreement with Modal including EU standard contractual clauses.

We do not release the speech sample or the voice model for training a general speech model. Under the processor agreement Modal may use the data only to provide the agreed speech synthesis, so training is not a permitted purpose. The agreement does not contain a separately worded training ban.

## How long the clone is kept

The voice clone stays stored until consent is withdrawn. The end of the subscription does not delete it. If the voice should not stay stored after that, withdraw consent as described below.

If you withdraw consent in the parent account or through the revocation link, the voice is blocked for new episodes straight away. A revocation by e-mail is carried out by support once they have checked it. With the block we delete any recordings and voice previews still stored at Heldenflug and order the deletion of the clone at the router. A revocation in the parent account sends that order immediately; otherwise the next daily deletion run sends it at the latest. If the order fails, the system retries it automatically. If it is still open after that, we follow it up by hand.

We delete the speech sample itself as soon as the clone has passed the quality check. If cloning fails, we delete it at the latest in the first daily deletion run after it is 24 hours old. It is not kept as a lasting backup of the clone.

## What happens on revocation

You can withdraw this consent at any time with effect for the future. You can do that:

1. through the delete or revoke function in the parent account,
2. through the personal revocation link of the voice person (to copy and print, without an account), or
3. by e-mail to [hallo@heldenflug.de](mailto:hallo@heldenflug.de). The voice person names the recording label and roughly when. Support checks this internally and does not tell the account holder who revoked.

After revocation:

  • the clone is no longer used for new episodes;
  • new productions with this voice are stopped if they are not yet finished;
  • from the next episode a released default voice is used;
  • the clone and any remaining speech samples are deleted as described under “How long the clone is kept”;
  • consent remains lawful for processing that already happened until revocation.

Episodes already published are not changed afterwards. They stay unchanged with the voice they were published with. Revocation does not mean those episodes are re-recorded or removed from the private feed with retroactive effect.

Revocation must not depend on giving a reason. Revocation ends use of the cloned voice; the audio-story subscription itself continues unless you cancel it separately. If an episode cannot be produced as planned without this voice, Heldenflug may finish the episode with a default voice or reschedule the affected episode under the subscription rules.

## Who gives consent

Please record a voice only if you are allowed to. Consent comes before the microphone and the first recording. It is not pre-filled.

With your confirmation you declare one of these roles:

  • I am the person whose voice is recorded (`self`): The recording contains my voice. I give consent myself — also if I speak on the account holder’s device in person, and also if I record through an invitation link without an account.
  • I have parental responsibility and I give consent for my child (`guardian_for_child`): I accompany the recording. A child does not consent themselves.

The earlier third-party declaration for adults present in person no longer applies to new consents. It remains effective only in existing `@v1` consents of legacy profiles.

Invited adults read a short check sentence before the first expression. It helps prevent someone playing another person’s voice. The sentence is processed only for this check.

Existing single profiles may still be used. Anyone who adds missing expressions gives this `@v2` consent for that. The old `@v1` consent covers only the previous everyday use.

If the voice person withdraws consent or disputes that they agreed, the voice is blocked and deleted under this text.

## Your express consent

With a checkbox that is not pre-ticked I confirm that I have read and understood this text. I expressly consent that Tiamat UG (haftungsbeschränkt) processes my speech recording, the voice features taken from it and a voice model created from it for producing and providing personalised Heldenflug audio-story episodes. I know that the clone can speak lines I never said. I know that episodes already published stay unchanged after a revocation.

☐ I give this consent expressly.

The role (own voice or child) is chosen separately and is not pre-filled.

Consent is voluntary. Without it that person cannot appear with their cloned voice in new episodes. The subscription can be used with default voices. Withdrawal is as easy as giving consent.