TTS Arena launches crowd-sourced benchmarking leaderboard for TTS models

AI Impact Summary

A benchmark platform enables side-by-side evaluation of TTS models by letting users submit text, listen to two models, and vote on which sounds more natural. The results feed a public Elo-style leaderboard with model names revealed only after voting to reduce bias, though the ranking reflects crowdsourced subjective judgments. The initial model lineup includes ElevenLabs, MetaVoice, OpenVoice, Pheme, WhisperSpeech, and XTTS, offering a mix of proprietary and open-source voices for cross-model comparison and competitive benchmarking.