whatmodel.sh / tts / OpenAudio S1 Mini

OpenAudio S1 Mini

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

OpenAudio S1 Mini is an open-weights text-to-speech model from Fish Audio, released June 2025. Its published size is 0.5B parameters. Licence: cc-by-nc-sa-4.0.

40.6±5.1#84 of 118 text-to-speech modelsfrom 4 benchmark results across 3 sources

Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.

$ how it compares

The text-to-speech models ranked either side of OpenAudio S1 Mini, out of 148 with a score, then more from Fish Audio.

#modelscorewhereRAMreleased
105Sonic (1) Cartesia42±3comparison only2024-05
106Nabra-82M v0.1 oddadmix42±10local~0.3GB2026-07
107OpenAudio S1 Mini Fish Audio41±5api price unknown +local~1.7GB2025-06
108Fish Speech 1.5 Fish Audio41±3local~1.5GB2024-12
109EMA Lightning EMA Lightning (canberkkkkkk)40±3local~0GB2026-10
more from Fish Audio
5OpenAudio S1 Fish Audio71±8api $15.00/M chars2025-06
31Fish Audio S2.1 Pro Fish Audio63±3api $15.00/M chars2026-06
37Fish Audio S2 Pro Fish Audio63±1api $15.00/M chars +local~11GB2026-03

see all text-to-speech models · find the ones that fit your machine

$ what it's good at

other tests4 not in a task · 90% of its score
testresultvs the others in its rundifficultycountswhy
LEMAS multilingual TTS eval · 9-language mean WER12.3% ✎█········█·█ beats 0/2avg 5.9%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 58%small run
Arabic TTS Arena · community blind A/B rating, 2026-09-24 snapshot931.4 ✎▃▂··▃·█▅···▂ beats 0/12avg 1018▇▇▇▇▇▇▇ 17%blunt
CodecMOS-Accent VCTK voice clone · human speaker-similarity MOS (S-SPK-SIM)4.16 / 5 ✎▄▄█·▄▄▄▄▄·▄█ beats 6/11easy · avg 4.20 / 5▇▇▇ 7.6%blunt
CodecMOS-Accent VCTK voice clone · human naturalness MOS (S-NAT)4.29 / 5 ✎█··▄▄██▄·▄▄▄ beats 10/11easy · avg 3.99 / 5▇▇▇ 7.0%blunt · at its level

$ where to run it

Through an API

api · price unknown · docs.fish.audio ↗

Other ways to run it

version

  1. Run it as published · needs ~1.7 GB, NVIDIA only · from github.com/fishaudio (official)
    git clone https://github.com/fishaudio/fish-speech && cd fish-speech && git checkout d3df50503b36314a964f66cac1af1e19e95bcfa3 && pip install -e ".[cu129]" && hf download fishaudio/openaudio-s1-mini --local-dir checkpoints/openaudio-s1-mini && python fish_speech/models/text2semantic/inference.py --text "Hello world" --compile && python fish_speech/models/dac/inference.py -i temp/codes_0.npy

Official weights and documentation: huggingface.co/fishaudio ↗

$ how the score was worked out · LEMAS multilingual audio suit… 58%, Arabic TTS Arena (Navid AI) ·… 17%, CodecMOS-Accent: listening-te… 15%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
LEMAS multilingual audio suite report1▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 58%CC BY-SA 4.0
Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-241▇▇▇▇▇ 17%Apache-2.0
CodecMOS-Accent: listening-test MOS of zero-shot TTS across English accents (arXiv:2603.14328v1)2▇▇▇▇ 15%CC BY 4.0
prior · pull toward the middle▇▇▇ 10%
weighted average= 40.6

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json