Radar ·
Open TTS Leaderboard adds 31M and 35B models and the UTMOS metric
The open text-to-speech leaderboard now includes a 31 million parameter model and a 35 billion parameter model, with similar English WER, plus a new naturalness metric called UTMOS.
The Open TTS Leaderboard, a panel that compares open text-to-speech models, now includes two models of very different sizes: phonon, from Gradium, with 31 million parameters, and Drama-3, from FishAudio, with 35 billion. According to the source, the two show similar English WER, 1.08 versus 1.19. The update also adds the UTMOS metric, which assesses the naturalness of English speech output.
The source summarizes that the panel shows how small and large open models compare on both WER and the new naturalness metric. The original text does not detail the test set, methodology or the cut-off date of the results. To verify, consult the leaderboard's official page and its metric documentation, and confirm the numbers before citing them. The date recorded in the source is 23 October 2026.
Original source: Open TTS Leaderboard adiciona modelos de 31M e 35B e a métrica UTMOS