Local Translation Models for Voxt
Voxt treats translation as its own model channel. That lets you translate spoken input after transcription or translate selected text without forcing the same model used for dictation and cleanup to handle every language task.
The short version
- Hy-MT2 1.8B Q4_K_M is the documented lower-latency local translation choice.
- Hy-MT2 1.8B Q8_0 is the documented higher-quality comparison choice.
- Local translation is useful when you want the translation step to remain on the Mac after the model is installed.
- Translation quality depends on language pair, terminology, source text, model variant, and hardware.
Model map
| Model or family | Role | Useful when | Upstream source |
|---|---|---|---|
| Hy-MT2 1.8B Q4_K_M The documented lower-size quantized translation variant for a lighter local workflow. | Local translation | Start here when responsiveness and a smaller local footprint are priorities. | Hunyuan Translation models |
| Hy-MT2 1.8B Q8_0 The documented higher-precision translation variant for quality comparisons. | Local translation | Compare it with Q4 when terminology and output quality matter more than model size. | Hunyuan Translation models |
Voice translation and selected text
A useful translation shortcut should work in the context where the text already exists. Voxt can translate spoken input after you finish speaking, and it can translate selected text directly. The best model choice can differ between a short message and a longer paragraph with product terminology.
Q4 versus Q8
Quantization is a practical trade-off, not a universal quality label. Q4 is a reasonable starting point when you want a lighter local model; Q8 is a useful comparison when you have the memory and want to test whether additional precision helps your language pairs and terminology.
FAQ
Which local translation model does Voxt support?
The current local translation documentation highlights Hy-MT2 1.8B in Q4_K_M and Q8_0 variants, with an additional compatibility variant. The exact model picker and availability can change with Voxt releases.
Can Voxt translate speech offline?
It can when the speech recognition and translation channels are both local, the required models are installed, and the workflow does not select a remote provider. Local model use still depends on Mac hardware and the supported language pair.