Course Workshop & Avatars
Avatars & voices
Your own figures from a photo and your own voices from a recording — with consent as a condition and one fixed rule: every face speaks only with its own voice.
Why#
A course works when the presenter is familiar. Universities build their figures themselves, without the provider's involvement — and nobody ever has a stranger's voice put into their mouth.
How it works#
Click pathLearning workshopAvatar workshop
- 1Open “Avatar workshop”.
- 20 · Record consent: name of the person depicted, date, period of validity, what it applies to (image, voice) and upload the signed proof (PDF, PNG or JPEG). Without proof nothing is recorded (Fig. 1).
Fig. 1 “Record consent”: person, date, period of validity, purpose and the signed proof. - 3Upload a photo; from it the figure is created with mouth shapes, blinking and eyebrows, then a running preview (Fig. 2).
Fig. 2 The photo for the figure, linked to a recorded consent. - 4Voice: record or upload a speech sample; duration, sampling rate and proportion of silence are checked before building (Fig. 3).
Fig. 3 The voice: record or upload a speech sample; duration, sampling rate and silence are checked before building. - 5The administration releases the figure for students. After that it can be chosen in every course workshop under “Figure & voice”.
Settings and switches#
- Switchable per tenantDefault: onModule: Eigene Avatare
- Module “Own avatars” (requires “Avatar lecturer”); building on the GPU and the voice are counted as usage.
The rule: every face has one voice#
A figure speaks with its own voice — the voice assigned to it when it was created — or not at all. There is no substitute voice that steps in when its own is not available at the moment. In the past such a fallback sounded like “it works” and was the opposite: a face with another person's voice.
The one exception is none: a course without its own cast is presented by the default figure, and it speaks with its own voice assigned to it.
This rule is recorded in the code and protected by automatic guards; a term that would promise a “substitute voice” does not reach the screen in any of the fifteen languages.
Consent as a technical condition#
- Image and voice are consented to separately.
- No figure is built without a stored, signed proof.
- A withdrawal takes effect immediately.
- Everything a figure says is labelled as AI-generated for learners.
Pre-produced and live#
The presentation of a slide is available as a pre-produced video (audio from the file). For follow-up questions a lip-synchronous head answers live. Both run with our own models on rented graphics servers.
Tips#
- A withdrawal of consent takes effect immediately: the figure no longer speaks.
- A figure without a voice does not speak. If the voice is missing, the answer appears as text, and the log says why.
Related features#
- My avatarLecturers make themselves the speaker of their courses: consent, photo or video selfie, voice, test, use — revocable at any time.
- Phase 3 · Speakers & InteractionWho speaks the course, in which languages, with which voice — and where the speaker stands on every slide.
- AI on the platform — what it may and may not doAI writes suggestions, presents courses and answers questions — labelled, bound to sources and with fixed limits: it does not assess exam behaviour and decides nothing that a person must decide.
Background knowledge
- Turning a presentation into a video with a speakerTurn slides into a video with voice and a visible speaker: your own recording, a synthetic voice or an avatar, placement, captions and MP4 export.
- AI avatars as lecturers: possibilities and limitsWhat an AI avatar can do as a lecturer, how pre-produced and live-answering avatars work and which rules apply to labelling and consent.
- AI voices in learning: speech synthesis and consentHow synthetic voices and voice clones work in courses, when consent is required and what the GDPR and the EU AI Act demand.