Put yourself in the commentator’s seat. The match is in front of you, the stats are on a screen beside you, and the producer is in your ear. You already know the players’ seasons, the rivalry and what this result would mean for the table. Every few seconds you choose: say something now, save it for later, or let the crowd carry the moment. When you do speak, your voice has to fit the moment, and the call has to land as the ball hits the net, not after.

An AI commentator has to do the same job, in real time, with models for eyes and data for ears. It turns partner feeds, sensors and video signals into one clear picture of what just happened. It works with the same background knowledge a commentator carries in their head. It picks the one thing worth saying, or decides to stay quiet, and remembers everything it has already said. It says the line in a voice that sounds real and carries the right emotion. And all of it has to happen fast enough for the call to arrive with the action.

ValkaAI Commentator splits this work into five stages, one for each of those jobs: an event feed, a context layer, the Sport Harness, ValkaTTS, and synchronization. Each stage passes a cleaner, more selective signal to the next, so the stages below follow the commentator’s own thinking in order.

› 01

Watch the match, and turn it into one event feed

A commentator works from what they can see, and our ValkaAI Commentator is no different. Its video understanding models watch the broadcast frame by frame and read the match out of it: where play is, who is involved, what happened and how. This is the input that depends on nobody else, and it is the one that decides how far the product can reach. Data deals reach a handful of top leagues. Cameras reach every match that is broadcast.

Fans read the game just by watching it. So can we.

But video can’t cover everything. Structured data supplied by partners is exact about what a camera cannot settle on its own: official times, decisions, sanctions. Purpose-built sensors add what neither can see. Each source arrives in its own format and at its own pace, so the first stage cleans and standardizes what arrives, then combines it into a single event feed.

From here on, every signal is based on one consistent picture of the match, not on the quirks of each provider. It also makes the system easier to extend: adding a source, a competition, or a sport does not require rebuilding every component that follows it. A new league or tournament with no data partner can go live on video alone and add other data sources later.

› 02

Add the story behind the action

Live data can tell the system that a goal was scored, a point was won, or possession changed. On its own, it cannot explain why that moment matters.

ValkaAI is building a contextual knowledge base containing player statistics, team performance, historical results, ongoing storylines, and relevant developments beyond the field of play. This gives the commentator material that the live event feed alone cannot provide.

The aim is not to fill every pause with another statistic. It is to make the right context available at the moment it becomes useful. A routine point may complete a player comeback. A goal may continue a pattern that has developed throughout the match. A result may carry consequences for the wider competition. Connecting those details is what turns a sequence of events into a story.

› 03

Decide what is worth saying

At the center of the system is the Sport Harness. Its job is not simply to produce a sentence. It has to decide whether the commentator should speak at all, and what deserves to be said when many possible talking points compete for attention.

It can react to the latest event, continue an existing storyline, introduce a new observation, save a useful detail for later, or leave space for the action itself. It also remembers what has already been said, helping the commentary remain consistent and avoid unnecessary repetition throughout the match.

› 04

Give the words the right voice

Once the system has decided what to say, ValkaTTS, ValkaAI’s proprietary text-to-speech system, turns the text into speech in real time.

For live commentary, sounding natural is only part of the problem. The delivery also has to match the significance of the moment. A missed opportunity may call for disbelief, a difficult period for restraint, and a last-second winner for an immediate burst of excitement. The same level of intensity cannot be applied to all three.

ValkaTTS provides granular control over emotional delivery, allowing the voice to follow the changing rhythm of the match. The goal is not constant excitement, but proportionate expression: enough emotion to make an important moment feel important without turning every line into a climax. All of it live. All of it in real time.

› 05

Keep the commentary connected to the action

The final stage synchronizes the generated audio with the live event or video feed. Timing is not a finishing detail. It determines whether the commentary feels like part of the experience.

Every stage of the pipeline takes time, from receiving and processing an event to choosing a response, generating the words, and producing the audio. If a goal call arrives after the viewer has already moved on, the sentence may still be correct, but it no longer feels live.

Synchronization keeps immediate reactions close to the event that triggered them and places broader observations into moments where the pace of the match allows them to breathe.