04 June 2026

Language Is Entertainment's Last Mile

Watching sport should be a pleasure. You sit down to relax, and the experience works best when it speaks to you in your own language, in a voice that knows the game. That simple truth is the one the international sports business has spent decades working around rather than solving. It is not only a sports problem. It is entertainment, and the businesses that crack it first will reach audiences that the rest cannot.

For years, the assumption among Western federations was that English would carry the day. It is the most widely understood broadcast language, so the world feed went out in English, and the job of localising it was handed down to individual broadcasters in each territory. That worked while budgets were generous and rights fees only went up. It works less well now. Budgets are tightening, federations are looking more closely at the markets they under-serve, and the global deals they sign only pay off if the content actually lands in each territory. Apple and MLS are the obvious example: a worldwide distribution deal is only worth what it earns in each market, and a market you cannot speak to is a market you are leaving on the table.

The technology has already solved most of the last mile. Virtual production localises the advertising and sponsorship, and you can localise on-screen graphics as well. Captions can be translated cheaply. Language is the final step and by far the hardest, because it is not a single problem. Arabic carries a range of dialects. Spanish splits into European, Central American and South American. Getting it wrong does not just miss; it can offend.

Where Thee Idea Came From

I have lived this problem from the inside. At Fox Sports in Asia, we owned major rights across a wide portfolio and delivered international feeds in multiple languages, including Formula 1, MotoGP, tennis, golf, badminton, and football. F1 is the example that stays with me. It went out in Mandarin, Cantonese, Bahasa Malaysia, Bahasa Indonesia, Thai and Vietnamese, alongside the English feed. The trouble was that the local commentators we hired often did not know the sport deeply, and viewers noticed. We would get complaints asking us to put the English feed back, because fans knew that David Croft and Martin Brundle understood F1 inside out. They would rather listen to an authority in a second language than to someone who did not know the subject in their first language.

That is when the thought first occurred to me. What if we could take Croft and Brundle, the voices that actually knew the sport, and render their commentary into the languages where no expert equivalent existed? My instinct was that a viewer would forgive the occasional mispronunciation if the substance was right and the voice carried genuine authority. The quality of the knowledge matters more than the perfection of the accent. At SailGP, years later, I hit the same wall from the other side, delivering promotional content into French, Spanish, Japanese, Mandarin and English, with more languages added as the league's roster of national teams grew.

The point is not to displace established commentary. Germany has superb F1 commentators and a deep, knowledgeable audience; you would never overwrite that, and you would be foolish to try. The opportunity lies in emerging markets, the territories where the sport is still niche, where the fans who do follow it are passionate and well-informed, but where the audience is too small to justify hiring a dedicated expert. Those markets get underserved or not served at all. That is where an authoritative voice, rendered into the local language, adds something rather than taking something away.

The idea sat with me until the technology caught up. I came across AI speech translation at Sportel in Monaco in 2024 and have followed it closely since. It is a fast-expanding market with the usual sports-tech arms race underway and a crowded field of competitors chasing the same prize.

There is an honest objection to meet here, and the same distinction answers it. Cloning a commentator's voice raises the obvious question of authenticity, but dubbing has been a paid craft for a century, and nobody calls it deceptive. Think of film. A German audience going to a Tom Cruise movie expects the familiar dubbed voice they have heard for years; that familiarity is the value, and you would not disturb it. In an established market, the existing voice wins. In an emerging market, where there is no established voice and no expert to hire, a cloned authority is not replacing anyone. It is opening a door that was closed. Handled properly, with the original talent consenting and retaining ownership of their voice, this turns a one-off commentary fee into licensable IP that works across every market the content reaches. The voice becomes an asset the talent controls, not a job they lose.

What the Test Taught Me

I came across Speechlab ( https://www.speechlab.ai/) in 2025, and we trialled it at This Sporting Planet, taking a podcast episode into Spanish, Vietnamese, Arabic and Mandarin. Then I did what the marketing decks never let you do: I used my own network to check the results honestly. For most languages, the verdict was that it still sounded, in their words, very AI-like. The Spanish, though, was excellent.

The reason mattered more than the result. Speechlab keeps a human in the loop, and much of its development team lives and works in Argentina, so the Spanish has real native judgment behind it. That is the insight the rest of the market keeps missing. The company has since done serious work in Arabic, bringing in a professor of Arabic to guide the translation, and the results were strong enough that the programming was approved for distribution in Qatar. Human review is not a weakness to be automated away as fast as possible. It is the thing that makes the output trustworthy, and Speechlab has built its process around accepting that, where others have rushed past it.

The Honest Answer on Live

Every federation I speak to asks the same question first: can it do live? The honest answer is not yet, not properly. The models can translate in near real time, but the latency is still too long. Speechlab's testing has brought it down to a 10- to 12-second delay, ahead of much of the field, which sits closer to 17- to 20-seconds. Neither is viable for a live signal. Get it under five seconds, and the conversation changes. That number will keep falling, and when it crosses the threshold, live localisation becomes real. We are not there in 2026. We are closer than most people think.

The Part That Matters Most

If there is one thing for federations and broadcasters to understand, it is this: the system learns. Every correction a human reviewer makes teaches the model, and over time, the number of mistakes falls. That is the whole point of the approach, and it is also where most organisations lose their nerve. They trial a tool, see the early imperfections, and walk away before the learning has had a chance to compound. The ones who will win are the ones patient enough to let the machine improve on its own content, in its own languages, with their own corrections feeding back in.

This is where the technology fits best today, and where the smartest investment sits. Speechlab is ideal for recorded content: highlights, previews, promos, and other short-form content, as well as classic matches sitting in the archive. So my advice to federations and broadcasters is direct. Use your archive to teach the machine. Imagine giving the system twenty years of footage to learn from. It would absorb the phraseology of the sport, how a save or a pass or an overtake is described, and how that language shifts from one market to the next. Yes, it takes a budget. But spend it here, and you are not buying a translation; you are building the asset that makes every future translation better. That could be the competitive advantage that carries your sport or your brand into a broader audience, and a broader audience brings new markets, and new markets bring the sponsors and brands that follow them.

The hardest part of that learning is not vocabulary. It is an expression. A commentator describes a goal with phrases that are understood by their audience but which may have no clean equivalent in English; every language carries its own colour. Teaching a model to localise the feel of the commentary, not just the words, is the real frontier. It is also where the difference between adequate and authentic will eventually be decided.

For now, the lesson from my own test stands. The technology is good enough today to serve markets you are currently ignoring, in voices your audience already trusts, at a cost that makes the long tail worth reaching. It will only get better, and it will get better faster for whoever commits to it first.

At LCA, this is the kind of decision we help businesses get right: not whether the technology works, but where in the commercial model it earns its place, and how to build the patience and the process to capture the advantage before the market closes it.


Disclosure: I sit on the advisory board of Speechlab. I have used the product in my own work, and the above is based on that, not on theory.