What does the sound engineer need for an AI avatar?
In short
Your sound engineer provides the moderator’s microphone as a separate channel and an aux send with the other stage microphones, pre-fader and without the avatar return. Add a return path for its voice and a screen or monitor for its picture. We bring the laptop with audio interface and the tablet and run them ourselves. Internet is a must.
Last updated 2 October 2026
01Echo
Keep the avatar return off the aux send
At the desk, two paths run to the AI avatar and one runs back. The moderator’s microphone arrives on a separate channel, the other stage microphones on an aux send, and a return path carries its voice into the PA. If the return also sits on the send, it hears itself. Every sentence it says then ends up in speech recognition, among the guests’ voices it is supposed to be listening to.
In one of our transcripts, more than half of the lines with no recognized speaker were its own. The laptop’s built-in microphone had picked its voice up again from the speaker, and speech recognition logged it as an extra speaker. None of those lines triggered anything.
On the laptop, that echo traveled through the air. At the desk it would travel down the wire. One move at the desk prevents that: the return stays off the send. That does nothing about echo through the air. Whether the stage microphones pick up its voice from the house speakers depends on the room.
02Rider
What your sound engineer provides, what we bring
Your sound engineer provides
- The aux send pre-fader, so it still hears the stage when you pull a stage microphone down for the room.
- A return path into the desk, so the room hears it through the PA.
- A screen or monitor where people can see it speak.
- A separate microphone for the moderator, on its own channel to our laptop. Then nobody has to assign that voice.
We bring
- A laptop with an audio interface where your send and the moderator’s channel come in and the return goes out.
- The moderator’s tablet and a hotspot that gets it online.
- A voice sample before the event, so it recognizes its name.
- One of us, sitting next to your sound engineer and running our equipment.
03Speakers
Who is speaking? Assign, but not too early
The guests arrive mixed on the aux send. By default, only their stage microphones are on it, so the avatar does not hear voices from the audience. If your sound engineer adds a room microphone, questions from the floor end up in the transcript as well. Before the event, we enter the names of the moderator and the guests on the tablet. It addresses them with those names, and in our live tests it called the guests by their correct names. Matching each guest’s voice to their name is a separate step we do once.
Assigning too early goes wrong. During one live test, two people spoke into the same laptop microphone on the first attempt, and speech recognition took their voices for one. The assignment then turned both into the moderator. We abandoned that attempt. A voice can only be assigned once it has spoken a few sentences.
In the second attempt, only the guest was assigned. The unnamed lines that were not echo then belonged almost entirely to the moderator. On a laptop, everyone shares one microphone. On stage, the moderator has their own, which is why it is on the rider.
04Internet
The tablet has its own network
Speech recognition and the language model run in the cloud. That makes the internet line part of the signal chain, just like the aux send.
That is why the moderator’s tablet runs on our hotspot rather than the venue Wi-Fi. Which line the laptop uses to get online is not settled by that. We decide it with your sound engineer.
05Coordination
We settle connectors and levels with your sound engineer
Connectors, levels and channels between your desk and our audio interface depend on your setup, so they are not on any spec sheet. We work them out directly with your sound engineer once the character and the date are set, and that is what the two weeks or so of lead time are for.
At the dress rehearsal, we then listen to the whole path through your desk together, including whether its voice finds its way from the house speakers back into the stage microphones.