▶TTS Arena
small models · blind votesARENA / TEXT-TO-SPEECH
Which voice
sounds better?
Blind battles between small text-to-speech models. Two read the same text, you pick the better one.
max 300 characters
MODEL A
MODEL B
Press Battle to hear two random models.
10 models · 10 rated battles
| # | Model | Elo | Params (M) | Elo / M | Battles | W | L | T | Both bad | Win rate |
|---|---|---|---|---|---|---|---|---|---|---|
| 01 | Kokoro 82M ↗ hexgrad | 1072.3 | 81.8 | 13 | 3 | 2 | 0 | 0 | 1 | 100% |
| 02 | cirxinx anonymous | 1038.7 | 13.1 | 79 | 2 | 2 | 0 | 0 | 0 | 100% |
| 03 | Phonon 31M ↗ Gradium | 1028.1 | 31.0 | 33 | 1 | 1 | 0 | 0 | 0 | 100% |
| 04 | tiger-alpha anonymous | 1024.5 | 1.7 | 602 | 2 | 1 | 1 | 0 | 0 | 50% |
| 05 | Piper (lessac medium) ↗ Rhasspy / OHF | 1000.2 | 15.7 | 64 | 3 | 2 | 0 | 0 | 1 | 100% |
| 06 | Kitten TTS Nano 0.8 ↗ KittenML | 998.4 | 14.0 | 71 | 0 | 0 | 0 | 0 | 0 | — |
| 07 | SupraTTS-1.0-Preview ↗ SupraLabs | 964.8 | 1.0 | 935 | 1 | 0 | 1 | 0 | 0 | 0% |
| 08 | Inflect Micro v2 ↗ Owen Song | 962.5 | 9.3 | 103 | 4 | 0 | 3 | 0 | 1 | 0% |
| 09 | BananaMind TTS V2 ↗ Banaxi-Tech | 692.4 | 13.1 | 53 | 2 | 0 | 2 | 0 | 0 | 0% |
| 10 | BananaMind TTS V1 ↗ Banaxi-Tech | 426.4 | 9.5 | 45 | 2 | 0 | 1 | 0 | 1 | 0% |
How it works
Type a sentence (or draw one), and two of the 10 models below are picked at random to read it. Every clip is loudness-matched before playback, so nobody wins by being louder. Each model reads with one fixed voice. You vote, then the names are revealed.Rating
Ratings are Elo (400 points per factor of 10 in odds), replayed vote by vote. Each model starts from a rating set by its automatic benchmark score (TTS Index: WER, UTMOS, SCOREQ, DNSMOS). How much a vote moves a rating depends on the gap: beating a stronger model gains more than beating a weaker one. Each model also has its own step size, large for its first battles and shrinking as it plays more, so two models rarely move by the same amount. Both bad counts as a draw but is tallied separately.Contenders
tiger-alpha 1.70M · anonymouscirxinx 13.12M · anonymous
Kokoro 82M ↗ 81.81M · hexgrad
Phonon 31M ↗ 31.00M · Gradium
Inflect Micro v2 ↗ 9.34M · Owen Song
Kitten TTS Nano 0.8 ↗ 14.01M · KittenML
Piper (lessac medium) ↗ 15.65M · Rhasspy / OHF
SupraTTS-1.0-Preview ↗ 1.03M · SupraLabs
BananaMind TTS V2 ↗ 13.05M · Banaxi-Tech
BananaMind TTS V1 ↗ 9.49M · Banaxi-Tech
Codename models are unreleased checkpoints, rated blind like everyone else.
Logging
Every text you submit and every vote is written to the arena's Hugging Face bucket. Don't type personal or sensitive information.Texts and votes are logged to a Hugging Face bucket. Audio is generated on demand; no clip is stored.