whatmodel.sh / tts / NeuTTS Air

NeuTTS Air

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

NeuTTS Air is an open-weights text-to-speech model from Neuphonic, released October 2025. Its published size is 0.552B parameters. Licence: apache-2.0.

Catalogue identity notes

neuphonic/neutts-air (Qwen 0.5B-based voice-cloning TTS with the NeuCodec decoder); README: '~552m' parameters including embeddings. The Q4 GGUF backbone (neuphonic/neutts-air-q4-gguf) is Neuphonic's own quantisation of the same model. Weights Apache-2.0; the neutts code repository uses the NeuTTS Open License. CodecMOS-Accent cites the neuphonic/neutts repository without naming the backbone file.

36.2±10.0#97 of 118 text-to-speech modelsfrom 2 benchmark results across 1 source* thinly tested, treat the rank as provisional

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 NeuTTS Air, out of 148 with a score, then more from Neuphonic.

#modelscorewhereRAMreleased
120Eleven v4 ElevenLabs37±4api $80.00/M chars2026-09
121Amphion VALL-E v2 recipe (MLS) Amphion (CUHK-Shenzhen)36±3local~2.9GB2024-06
122NeuTTS Air Neuphonic36±10local~4GB2025-10
123Eleven v4 Turbo ElevenLabs36±4api $40.00/M chars2026-09
124VITS (VCTK multi-speaker) Kakao Enterprise (Jaehyeon Kim et al.)34±5local~0.2GB2021-06
more from Neuphonic
75Neuphonic neu_hq Neuphonic52±3comparison only2025-08

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

$ what it's good at

other tests2 not in a task · 60% of its score
testresultvs the others in its rundifficultycountswhy
CodecMOS-Accent VCTK voice clone · human speaker-similarity MOS (S-SPK-SIM)3.27 / 5 ✎▄▄█·▄▄▄▄▄·▄█ beats 0/11easy · avg 4.20 / 5▇▇▇▇▇▇▇▇▇▇▇▇▇ 31%blunt · at its level
CodecMOS-Accent VCTK voice clone · human naturalness MOS (S-NAT)4.06 / 5 ✎█··▄▄██▄·▄▄▄ beats 8/11easy · avg 3.99 / 5▇▇▇▇▇▇▇▇▇▇▇ 29%blunt · at its level

$ where to run it

Other ways to run it

version

  1. Run it as published · needs ~4.01 GB · from huggingface.co/neuphonic (official)
    git clone https://github.com/neuphonic/neutts.git && cd neutts && pip install "neutts[llama]" && python -m examples.basic_example --input_text "My name is Dave, and um, I'm from London" --ref_audio samples/dave.wav --ref_text samples/dave.txt --backbone neuphonic/neutts-air-q4-gguf

Official weights and documentation: huggingface.co/neuphonic ↗

$ how the score was worked out · CodecMOS-Accent: listening-te… 60%

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
CodecMOS-Accent: listening-test MOS of zero-shot TTS across English accents (arXiv:2603.14328v1)2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 60%CC BY 4.0
prior · pull toward the middle▇▇▇▇▇▇▇▇▇▇▇▇ 40%
weighted average= 36.2

$ 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