AI avatar glossary for organizers and sound engineers

01Mixing desk

What the AI avatar hears at the mixing desk

The sound engineer settles at the mixing desk whether the AI avatar hears the guests or its own voice. The words for it come up in planning, often unexplained.

Tech rider

The written list of what the event crew provides. For an AI avatar: mainly the moderator’s own channel, the aux send, the return path and a screen.

Aux send

An extra output on the mixing desk with its own mix. Here it carries the stage microphones, except the moderator’s, to our laptop.

Pre-fader

The aux send picks up the signal before the channel fader. If the sound engineer pulls a microphone down for the room, the AI avatar still hears it at the same level.

Return path

The line carrying the AI avatar’s voice back into the mixing desk, so the room hears it through the speakers.

Echo

The AI avatar’s own voice coming back to it, transcribed as one more speaker. Keeping its return off the aux send keeps it off the line. It can still come through the air, from the room speakers into the stage mics.

02Technology

Between question and answer

These words come up when someone asks why the answer isn’t instant.

Speech recognition

It transcribes what is said on stage live, speakers included, for now in standard German only. The AI avatar thus knows the last few minutes word for word, the rest as a summary.

Pause detection

It decides whether someone has finished or is just taking a breath. Too impatient, and the AI avatar cuts people off.

Language model

The AI model that composes the AI avatar’s answer. It runs in the cloud: the transcribed text goes there to be turned into an answer, while the transcript itself is stored on our laptop.

Speech synthesis

Software that turns the composed answer into an audible voice.

Latency

The wait between question and answer. The moderator’s tablet displays it.

First audio

At Aura51, the latency from the last word of the name call to the first audible sound, a short “Also.” or “Hm.” before the content. Median: about two seconds, hardly ever under about 1.5 s.

03On stage

When it speaks in the room and when it doesn’t

These words matter most to the moderator, who calls on the AI avatar and can stop it.

Name call

The moderator names the AI avatar and asks it a question. Without such a call or a tap on the tablet, it says nothing, except to a follow-up in the conversation window.

Conversation window

At Aura51, a set 12 s after its last sentence in which the AI avatar takes follow-ups without its name; an “Und warum?” (And why?) is enough. A “Danke” (thanks) gets no reply, and handing the floor to someone else closes the window.

Hotspot

A separate wireless network that we bring along. The moderator’s tablet runs on it, apart from the venue Wi-Fi.

Answer preview

The answer as text on the moderator’s tablet, before the room hears its content.

STOP

The tablet button the moderator uses to cut off an answer. In live tests, the AI avatar went silent in under half a second.

Interjection

At Aura51, only its name, called by the moderator into its answer. That it then pauses once per answer, says a short line such as “Moment — …” (one moment) and resumes at the start of the interrupted clause, we have measured in the lab, not yet with people in the room.

04Terms

AI avatar, chatbot or digital human?

On stage, these words differ in whether the counterpart listens and who gives it its words.

AI avatar

A face and a voice from the computer, driven by an AI rather than a script or a person. It is called interactive when it responds live to what was just said.

Video avatar

An avatar whose video is made in advance from a script, say for a welcome message. Nothing said in the room afterward reaches it.

Chatbot

A program that answers what a single person enters, usually in writing.

Digital human

An industry term for a human-like, usually photorealistic computer-made likeness, whoever controls it.

05At events

Which character, with what brief

With the event set, talk turns to character and role.

Character

An AI avatar with its own name, face, voice and role. Aura51 has four. Albert plays the provocateur, Vera the skeptic, Tina explains, and Ken is the voice from outside.

Briefing

What an AI avatar needs to know before the event: topic, participants and the position it takes. The briefing decides whether it speaks for a product or plays the critic.

Devil’s advocate

Someone who argues the opposing side on purpose, so a decision gets tested. At an internal town hall, an AI avatar can play that part, one that often no one else takes on.

06Law and evidence

What the law requires and what we can prove

Labeling and deepfake come from the EU AI Act, evidence level from us.

Labeling

A notice that something is AI-generated, for example on the screen. Under Article 50 of the AI Act, people interacting with an AI must as a rule be told so since August 2, 2026; for the audience, a notice is good practice.

Deepfake

Under the AI Act, AI-generated or manipulated image, audio or video content that resembles something real, such as a person or place, and would falsely appear authentic. In our reading, an invented character lacks this feature.

Evidence level

At Aura51, how far a claim is backed, from “measured · live” to “built · not yet verified live”. “Live” means a laptop test with people in the room.

If a word comes up in your planning meeting that is missing here, ask us before anyone plugs anything into the desk.