What is a Quality Elo score for AI voices?
Quality Elo is a listener-vote rating for voice engines, borrowed from chess: blind pairwise comparisons produce a score nobody can self-award. How to read one, with the attribution.
Updated June 11, 2026
The definition
Elo is the rating system invented for chess: players gain or lose points by beating or losing to each other, and the running score ends up encoding strength across thousands of head-to-head results. A Quality Elo applies the same machinery to voice engines. Listeners in a public arena hear two clips of the same text, blind, and vote for the better one; every vote nudges the two engines' ratings, and a leaderboard emerges from the crowd rather than from any vendor.
The blind pairing is the point. Nobody can vote their own engine up without out-voting the public, which makes an arena Elo one of the few voice-quality numbers a vendor cannot tilt.
The attribution
Every Quality Elo printed on this site is third-party data from the Artificial Analysis Speech Arena, retrieved 2026-06-10. They are user-vote arena ratings, not our scores, and we print the retrieval date next to them because arena ratings move as votes come in.
How to read one
Differences matter more than absolute values. For scale: the top-rated model of all roughly 85 in the arena, Fun-Realtime-TTS, sits at 1228.06, and Gemini Flash, our highest-rated engine, sits at 1225.13, within a few points of it. A gap of a few Elo is statistical noise; a gap of a hundred is one engine winning most blind matchups.
Vote volume matters too. A rating built on thousands of votes is solid; a fresh entry near the 1,000 baseline mostly reflects how few matchups it has played. Honest leaderboards footnote this instead of hiding it.
Two live examples of the footnote rule on our own roster: MAI Voice 2 carries the rating of MAI-Voice-1 because version 2 is not yet arena-rated, and Zonos sits near the baseline with limited votes so far.
Where Cantari uses it
Quality Elo is the quality column of the open benchmark and the score on every engine card, printed with the attribution whether the number flatters us or not. Why a third-party score anchors our routing is the subject of the benchmark post, and the roster-level reading is in the engines guide.