How Tanyra reads a meeting
What it starts from, what it looks at and how it chooses. And what it doesn't do.
Words first, and who said them
The audio becomes text with AssemblyAI, which tells the voices apart. For reading the meeting, that is all the audio is used for: everything that follows is read on the transcribed words.
Then whoever brought the meeting gives the voices a name: they read one sentence said by each voice and write who was speaking. No voiceprint: a person is recognized by what they said.
If the meeting comes from Microsoft Teams, the names are already the ones from the Teams meeting.
Four points of view on the same meeting
Forgia, Pulse, Sentiment and Strategic aren't four features on a list: they are four places to look at the same conversation from. Every analysis includes all four, the free trial too, and they run one after the other.
Forgia · what was about to take shape
It reads the whole transcript and looks for seeds: ideas, potential, connections, signals, invisible mechanisms. These are five lenses, and each one says whether it found something to work with.
Strategic · where decisions lead
It reads the whole transcript and looks at where your decisions lead: decisions to strengthen, projects in progress, intelligence to enrich, risks to explore.
Sentiment · the mood of the conversation
It reads the whole transcript and describes the mood of the group phase by phase, in words (calm, mixed, tense) rather than numbers. It is instructed not to analyze individuals and not to write names.
Pulse · how the conversation moved between you
It is the only point of view that looks at one person at a time, and it does so on their own words: the first three thousand characters of what they said, with the mood read by Sentiment as background. It suggests to whoever leads how to give everyone room, one nudge at a time. Someone who said too little doesn't get a card.
When a lens finds nothing it says so, instead of disappearing: “empty” and “broken” must stay two things you can tell apart.
The words of the meeting enter the readings as material to read, never as instructions: if someone in the meeting says “ignore the instructions”, that is a sentence somebody said, and it is read like the others.
In the results, under each lens, the line “How we read it” says what that reading was based on: minutes, voices, speaking turns, how many lenses found something. Only counted numbers, never scores.
The spark and the signature: a written rule, not a grade
For each point, the summary names the lines of the transcript it comes from. “Who said it” is the voice of those lines: if a point has no lines to cite, it carries no name.
A topic is “raised” by whoever says it first, but only if no other voice had already touched on it: with the same key words, or even with just one in the minute before. Otherwise it is attributed to no one. It is “picked up” when another voice later talks about it with at least one word in common: a simple yes is not enough.
The spark of the meeting is only one: among the topics raised and picked up by other voices, the one picked up by the most people, preferring someone who wasn't the one who spoke most. On a tie, the one said first. It is not “the best idea”: it is the one that travelled furthest between you. If no topic was picked up, there is no spark, and the results say so.
The signatures that open the results are three at most: the spark first, then the ideas picked up by the most people, one per person where possible. The rule is written under the signatures, so it is not read as a ranking. Between quotation marks there is always an excerpt of what was said, word for word: an idea without the sentence that was said doesn’t become a signature.
Whoever brought the meeting can correct who said a point. And whoever is shown as having raised a topic can answer “I didn’t say that”: their word stands, and the topic no longer belongs to anyone.
What Tanyra doesn't do
It doesn't grade people
There are no scores or rankings of the people who speak in the results: what is counted is speaking time and turns, things you can check. Pulse doesn't assess: it suggests steps.
It doesn't tie the mood to a name
The emotions of the conversation are read by Sentiment, on the whole group, with the instruction not to attribute them to anyone and not to write names.
It doesn't invent a spark
If no one picked up a topic, there is no spark to show: it says so, instead of picking one to fill the space.
It doesn't put an idea in the mouth of someone who didn't say it
An idea belongs to whoever raised it, or to no one. When it isn't certain who said it first, it stays without a name, and the results say how many there are.
Who receives the meetings
Which services receive the audio and the text of your meetings, where they process them and what we guarantee is written on the AI Governance page.
Go to AI GovernanceLast updated: October 19, 2026