Open TTS Leaderboard: 31M phonon nearly matches 35B Drama-3 on English WER
realmrfakename · x · 2026-10-23
The Open TTS Leaderboard added two models: phonon by gradium (31M) and Drama-3 by Fish Audio (35B) — 1000x apart in size with similar English WER (1.08 vs 1.19). A new metric, UTMOS, now rates English output naturalness. Kokoro-82M still leads with 0.93 average WER, while Drama-3 ranks 11th overall but posts a strong 4.47 UTMOS.