24 AUG 2026
Glossary
Six terms. Each one is a word that means something different in a room with voices in it than it means in a text product, and that difference is the reason it is here. A term whose meaning is unchanged by speech is not on this page, however often it appears elsewhere on the site.
floor
The floor is the right to be the sole voice speaking in a Tingvar room, and exactly one persona holds it at a time.
The floor is held for a bounded period and then released, and it can be taken from whoever has it. The room grants it, never the persona: Gemini's server side voice activity detection is switched off, so no persona ever decides on its own to speak. Holding the floor is a lease with a time to live, and releasing it is a compare and swap, so two personas cannot both believe they hold it. Your voice takes the floor from whoever has it, which is what makes barge-in mean something rather than being a button that files a request.
| Not this | Why the two are different |
|---|---|
| a turn | A turn is a position in an order that was decided in advance. The floor is decided each time, from what has just been said. |
| a queue position | A queue is something you wait in until it reaches you. The floor can be taken from whoever currently holds it, including by you. |
| a microphone permission | Your microphone is open for the whole session. The floor is about who is being listened to, not about what is switched on. |
Never called turn, queue
arbiter
The arbiter is the room level policy that decides which persona takes the floor next.
The arbiter is not a persona. It has no opinion about your plan, it never speaks, and it is not a fifth voice settling the argument on your behalf. It is pure policy over events, with no network access and no audio, which is why it can be run and tested offline. Each persona bids to speak, the bids are scored locally against sentence embeddings rather than by a model call, and the arbiter grants the floor to one of them. Agent level turn-taking, where each agent decides for itself whether to speak, stops working at three agents.
| Not this | Why the two are different |
|---|---|
| a moderator | A moderator has a view on what is being said and steps in when it dislikes the content. The arbiter reads intent to speak, not argument. |
| a judge | A judge rules on who was right. Nothing in the room rules on your plan, and the transcript is left for you to read. |
| a synthesiser | A synthesiser writes the summary that resolves the disagreement. We publish the disagreement instead, because the disagreement is the product. |
Never called moderator, judge, synthesiser
barge-in
Barge-in is speaking over a persona so that it stops, and your voice takes the floor from whoever is holding it.
Human speech preempts everything in the room. That is an invariant of the design rather than a setting you switch on, and it is why speaking is how you steer a session at all. Speaking is also how you change the subject, ask a persona to defend a claim, or stop a line of argument that is going nowhere. Barge-in is not a raised hand, not a queued turn and not a pause button. The client leg of barge-in is not yet measured, and we publish measurements rather than targets.
| Not this | Why the two are different |
|---|---|
| a raised hand | A raised hand is a request that somebody else grants. Speaking takes the floor at once and does not wait to be recognised. |
| an interruption request | A request is a form submit that a system chooses to act on. Your voice stops the audio, and nothing decides whether to allow it. |
| a pause button | A pause stops everything and resumes where it stopped. Barge-in hands the floor to you and the room carries on from what you said. |
Never called raised hand, interruption
room
A room is the live audio space a session happens in: the personas, you, the floor, and the thing that decides who holds it.
You join a room from a browser. You do not open a chat with it, and you do not read it later as a thread, because by then it is over. A room is a place you are in, in real time, which is why join, in and leave are the right words for it everywhere on this site. What a room leaves behind is a transcript, and a transcript is a record of the room rather than the room itself. A saved conversation you can add to tomorrow is a different product, and several competitors sell it under this word.
| Not this | Why the two are different |
|---|---|
| a group call | A call connects people who each decide when to speak. Here the room decides, and three personas would otherwise answer every sentence at once. |
| a chat thread | A thread waits for you and keeps its place. A room runs in real time and the argument moves on whether or not you say anything. |
| a meeting recording | A recording is what is left afterwards. The room is the live thing, and the transcript is what you take away from it. |
Never called call, chat, meeting
turn-taking
Turn-taking is the decision about who speaks next, and in a Tingvar room the arbiter makes it once for the whole room.
In text, turn-taking is trivial: nothing overlaps and nobody is kept waiting. In speech it is the hard part, and it is where the latency budget is spent. Agent level turn-taking, where each agent decides for itself whether to speak, stops working at three agents, because three personas then answer every human utterance at once and the room is noise. Room level turn-taking can see every persona's intent before it grants the floor to one of them. It is the core of what we build, and it is the reason a room sounds like a conversation.
| Not this | Why the two are different |
|---|---|
| a speaking queue | A queue fixes the order before anyone has spoken. The decision here is made again after every utterance, on what was just said. |
| round robin ordering | Round robin gives everyone an equal share whether or not they have anything to add, which is how a room fills with filler. |
| a rate limit | A rate limit caps how often something may speak. Turn-taking chooses between several speakers who all want the floor at the same moment. |
Never called speaking queue, round robin
spatial attribution
Spatial attribution is knowing which persona said something because its voice comes from a fixed position in the stereo field.
Each persona's audio is panned to its own position, and that position does not move for the length of a session. Separating voices across the stereo field improves the ability to follow rapid speaker changes and to say afterwards who said what. It is a property of the audio rather than of the screen, so it holds when you listen with the display off. A mono fallback is offered, because many people listen on one earbud or a laptop speaker where panning does nothing and attenuating the far positions is actively harmful. Position is never the sole cue: the live transcript labels every line.
| Not this | Why the two are different |
|---|---|
| surround sound | Surround sound is about immersion in a scene. This is about telling four voices apart while they argue over each other. |
| 3D audio | Head tracked spatial formats move sound as you move. Here each position is fixed, because a moving voice is harder to attribute. |
| a speaker label on screen | A label works only while you are looking at it. The pan works while you are walking, driving or reading the brief again. |
Never called surround sound, 3D audio
Three terms defined somewhere else
These three are product vocabulary rather than category vocabulary, and speech does not change what any of them means. Each is defined in place, on the page where a reader first meets it, which is a better place for a definition than a list a reader has to go and find.
The mechanism these six words describe is one page long. Read how it works