Could You Play These Games By Ear?

This site has 28 games and not one byte of audio. No sound effects, no music, nothing that speaks. That was never a decision anyone took — it is just where a text-and-image site ends up. So I ran the obvious test on the catalogue: could you play these by ear? For 26 of them, yes. For 2, not in principle.

The test: say the prompt without saying the answer

That is the rule, and the cover chart above is the whole catalogue run through it — one dot per game, two crossed out. Most of it passes easily, and some of it passes better than the screen version does. “Who scored more league goals, Alan Shearer or Thierry Henry” is a complete round of Higher or Lower with nothing lost. A Futbol Grid board is three clubs and three countries, which is a sentence. Even Emoji Quiz survives, because emoji have names — “flag of Croatia, magic wand, first-place medal” is still a rebus, just a spoken one.

The failures are not awkward cases. They are structural.

Both of these hide information inside an image and reveal it gradually. Blur is not decoration there, it is the entire difficulty curve. You cannot narrate a partially-revealed face, because language does not have a blur setting — any description is either useless or complete. That is a real limit, not an implementation gap, and it is the honest answer to “why not just add narration to everything”.

Which leaves 26 games and one question

If 26 of our games could be played by ear, something has to do the talking. And the moment you ask what, you are not solving a technical problem any more.

A narrator here would have to get through 766 player names from 69 nationalities and 201 clubs. We have written before about what happens to those names when software flattens them, and a generic English voice reading “Szczęsny” makes exactly the same mistake in a different medium. Voices are not interchangeable at that scale; a roster this international needs more than one of them.

Whose voice, though

Cloning a voice is now about as hard as resizing an image, which means the interesting question stopped being “can we” some time ago. A synthetic voice reading a trivia question aloud represents nobody and claims nothing; it is the audio equivalent of a placeholder avatar, and there is no argument against it.

A named commentator’s voice is the opposite case, and football makes it unusually concrete. Commentary is one of the few jobs where the voice is the professional asset — people can identify a handful of commentators from three words. Cloning one to read our puzzles would not be a stylistic choice, it would be putting sentences in a working person’s mouth that they never said and were not paid for.

What makes a voice harder than a crest

We went through a version of this with club crests, and the conclusion transfers: an asset that identifies a real party cannot be synthesised, however good the synthesis is. But a voice is harder than a crest in two directions.

A crest belongs to an organisation, is registered, and everyone already agrees that copying one is forgery. A voice belongs to a person, has far thinner legal protection in most places, and until recently could not be copied at all — so the norms that exist for logos simply have not been written for it yet. The other direction is worse: a wrong crest is obviously wrong once you look it up, while a cloned voice saying something plausible is indistinguishable from a recording. There is no equivalent of “check the club’s website”.

None of which is an argument against the tools. Cloning your own voice, licensing one properly, or generating a voice that belongs to nobody are all straightforward uses, and for a multilingual roster like ours the ability to produce AI voice cloning across languages is the difference between one narrator mangling 69 nationalities and a set of voices that can each say their own names properly. The test is not whether the audio is synthetic. It is whether a listener would believe a specific real person said it.

How this was built

The zero-audio claim is checked rather than asserted: a test scans the source tree for audio elements, the Web Audio and speech-synthesis APIs, and any bundled sound file, and fails if one appears. That way this page cannot quietly become wrong the first time somebody adds a whistle.

The 2-of-28 split is a judgement, not a measurement, and it deserves saying plainly after four posts of derived figures. The rule is stated and you can apply it yourself: describe the prompt out loud, and if every accurate description contains the answer, the game cannot be audio. Two games meet that. Several others — the grid boards, the drag-to-order ones — would be clumsy by ear rather than impossible, and I have counted clumsy as passing.

Play with the same data

The two that failed are worth playing precisely because of what they do with an image: Guess the Badge and Guess the Footballer. For the other end of the catalogue, try Football Quiz — ten questions that would lose nothing at all if you heard them instead of read them.