AreaAudit

Which languages each generator says it supports, and where

Which regions this category has barely addressed

Read by language rather than by vendor, a column of declared sets shows where an industry has spent and where it has not. The gaps are not random: they cluster by region and by writing system. As of 2026-09-12.

Where the declarations run out, region by regionWestern Europe and East Asia are named by most declared sets. Southeast Asia is named by a few, always as a block. South Asia is named by one. No declared set in this atlas names an African language at all, and nothing published about voices fills any of those gaps.Best coveredWestern Europe and EastAsiaNamed by most declared sets. Voice lists rarely name them.Southeast AsiaA few sets, always naming three or more of its languages atonce.South AsiaOne set, one language. No voice disclosure at all.AfricaNo declared set names any of its languages.Where the evidence runs out entirely
Fig. 1 The pattern follows business software spending rather than audience size, which is what a column of marketing declarations should be expected to show.
Which lists each vendor actually namesFilled where the vendor names something in that column, hollow where no list is published at all.Which lists each vendor actually namesTypical number of declared sets naming its languagesTypical numberof declared set…Published about voicesPublished aboutvoicesWestern EuropeWestern Europe — Typical number of declared sets naming its languages: Most of themWestern Europe — Published about voices: RarelyEast AsiaEast Asia — Typical number of declared sets naming its languages: Most of themEast Asia — Published about voices: RarelySoutheast AsiaSoutheast Asia — Typical number of declared sets naming its languages: A fewSoutheast Asia — Published about voices: NothingSouth AsiaSouth Asia — Typical number of declared sets naming its languages: OneSouth Asia — Published about voices: NothingAfricaAfrica — Typical number of declared sets naming its languages: NoneAfrica — Published about voices: Only inside long undifferentiated lists
Fig. 2 Filled where the vendor names something in that column, hollow where no list is published at all.
How thoroughly each broad region is addressed by declared interface sets. Recorded 2026-09-12.
RegionTypical number of declared sets naming its languagesPublished about voices
Western EuropeMost of themRarely
East AsiaMost of themRarely
Southeast AsiaA fewNothing
South AsiaOneNothing
AfricaNoneOnly inside long undifferentiated lists

Inclusion rule. Broad regions, with how many declared interface sets name any of their languages. Counts are of declarations, not of capability. Order. Most thoroughly addressed region first.

1The pattern follows software spending

The best-covered regions are the ones that buy business software in their own language, not the ones with the most speakers or the largest video audiences. Several very large language communities appear in one declared set or none.

That is what a column of marketing declarations should be expected to show. It is a map of where revenue is anticipated, and reading it as a map of audiences produces exactly the wrong conclusions.

2Southeast Asia is entered as a block or not at all

Where a vendor addresses the region it usually declares three or more of its languages at once, and where it does not it declares none. Nobody picks one market there and skips the neighbours.

So a reader in that region has a short shortlist and a clear one. What they do not have is any published statement about whether a finished video would speak their language, because none of the vendors involved publishes one.

3South Asia and Africa are where the evidence runs out

One declared set reaches South Asia, with one language. No declared set names an African language at all. Both regions may well be served by products whose long voice lists are undifferentiated, and nothing published would show it.

That is a limit of a register built on declarations. It records where companies have paid to speak to buyers, and a region absent from that map is absent from the evidence rather than necessarily from the market.

4Writing systems track the gaps

The regions with the thinnest coverage are disproportionately the ones using scripts other than Latin and Chinese. That matters twice: once for the interface, and again for any text a model has to draw inside a picture.

Nothing in this category publishes which scripts it can render. So the regions least addressed by the interface column are also the ones where the least repairable failure is most likely.

An explainer about method and structure, not a count. No vendor number appears here that is not also in the register. The sourced material is on the register. Related: The document layer, Figures that never expire.