Arena
Two clips, same sentence, different TTS providers. Names hidden until you vote. Every clip is real API output synthesized for this arena - the reference texts are the same FLEURS transcripts the leaderboard measures. Rankings are Bradley-Terry Elo with bootstrap 95% intervals, recomputed from raw votes.
No published rankings yet - they appear after enough votes land. No estimates are shown in the meantime.
Elo is recomputed from every raw vote; nothing is hand-set. Field mean is 1500 by construction.
Rankings shift as votes come in. Leave an email and we send the updated arena report when it changes. Only used for that.