docs
The wizard picks a sensible one, and it can be changed at any time in Languages & models. What the choice actually decides:
Whisper Medium q5
Covers everything and translates by itself — into English properly, into the other interface languages experimentally. About 1.5 GB.
a narrow model for your language
GigaAM v3 turned an 11-second phrase into text in 0.47 s against Whisper's 11.6 s, in a third of the memory.
Nemotron 3.5 streaming
Partial text appears on the plate a few times a second, 40 languages.
Whisper Base or Moonshine
Small and quick; less accurate on long sentences.
The catalogue shows the memory each model will actually take, measured against what is free right now, and says plainly whether it can translate.
Models are downloaded on request, from the repositories below, and every file is compared against a published SHA-256 before it is used. None of them is bundled with the program.
The default choice for every language, and the models that carry translation: into English through their own translate task, into other languages by being told the speech is already in the target one. Quantised builds are marked q5.
A single-language model that punctuates by itself. On an 11-second phrase it took 0.47 s against Whisper Medium's 11.6 s, and 277 MB of memory against 814 MB.
The previous generation of the same model: the same speed, without the punctuation.
25 European languages in one narrow model, punctuates by itself.
Types as you speak: partial text on the plate, 40 languages, punctuation included.
English, German, Spanish and French, and it translates between them without Whisper.
About 30 languages, the heaviest and the most accurate in the catalogue.
Small and quick, for one language at a time on modest hardware.
Post-processing runs whatever GGUF model you point it at, searched and downloaded from Hugging Face inside the program.
The catalogue is not a fence. Drop a Whisper model file (ggml-*.bin) or a sherpa-onnx model folder into the models folder next to the program, and it appears in the list under its own name, marked as found locally. The same goes for the editor: post-processing runs any GGUF file you point it at, and the search inside the program fetches one from Hugging Face.
Whisper translates into English with its own translate task, and into the other interface languages by being told the speech is already in the target one. Canary translates between English, German, Spanish and French by itself. The narrow models transcribe only: when one of them serves your language, the program keeps the best installed Whisper standing by, and the plate offers to hand a phrase to it.
Where each language's model is set, and how the rows work, is in the Languages and models guide.