Supertonic (on-device, ONNX, 99M)

Lightning-fast on-device multilingual TTS. 99M-parameter ONNX model, ~400 MB download, 31 languages, 10 preset voices. No GPU required.

Models

Model	Download
supertonic-3	400 MB

Voices

10 voices in 0 languages.

Options

In Settings › Add-ons › Supertonic (on-device, ONNX, 99M) › Configure.

Model

Default: supertonic-3

Larger model = broader language coverage, longer first-run download. Switching reloads the model on the next synthesis call.

Inference device

Default: cpu

Quality (sampling steps)

Default: 8

Higher = cleaner audio, slower. 5 is fastest preview, 8 is the upstream default, 12 is the quality ceiling. Past 12 the model saturates — additional steps cost compute without measurable improvement.

Language hint

Default: auto

Custom voice JSON (optional)

Default: empty

Path to a Voice Builder export (https://supertonic.supertone.ai/voice_builder). When set, the addon exposes one additional voice id 'supertonic-custom' that loads this JSON instead of a preset. The 10 preset voices remain available.

Languages

ar
bg
hr
cs
da
nl
en
et
fi
fr
de
el
hi
hu
id
it
ja
ko
lv
lt
pl
pt
ro
ru
sk
sl
es
sv
tr
uk
vi
na

Platforms

System	Download
Linux x86-64	78.1 MB
macOS (universal)	49.1 MB
Windows x86-64	45.6 MB

License

Add-on

MIT (wrapper) + MIT (supertonic-py) + OpenRAIL-M (Supertonic-3 weights)