docs
Choose a speech model for your language
The wizard picks a sensible one, and it can be changed at any time in Languages & models. What the choice actually decides:
Updated · Vitalii Yemets
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.
What the catalogue offers
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.
Which model for which language
One short answer per language: which model to pick, what you gain, and what to do on a weak machine. The catalogue above has the details of each model.
English
English is the language every model in the catalogue was trained on most, so the choice is about speed and what else you need. For words that appear while you are still speaking, take Nemotron 3.5: it streams partial text to the plate and punctuates by itself. For the lowest latency after you let go, Parakeet TDT 0.6B v3 turns a sentence into text in well under a second on a CPU and adds punctuation. Whisper Medium q5 is the safe default if you also dictate in other languages or want translation, at the cost of speed and about 1.5 GB of memory. On an old laptop, Whisper Base or Small or Canary 180M keep the program light; Canary also translates between English, German, Spanish and French without Whisper. Qwen3 ASR is the heaviest and the most accurate option when memory is not a concern.
Ukrainian
For Ukrainian the program recommends Parakeet TDT 0.6B v3: a narrow model for 25 European languages that punctuates by itself and answers in a fraction of the time Whisper needs. Nemotron 3.5 adds live text on the plate while you speak and lists Ukrainian among its languages; if a phrase comes out wrong, go back to Parakeet. Whisper Medium q5 stays the general choice when you also want translation into English or another interface language, and the program keeps the best installed Whisper standing by for that even when Parakeet serves the language. On a weak machine, Moonshine Base uk is a small single-language model that runs on modest hardware; it is downloaded by hand from its catalogue row. Qwen3 ASR covers Ukrainian too and is the most accurate of all, but it is the heaviest in the catalogue and needs a computer with room to spare.
Russian
Russian has a model of its own: GigaAM v3 recognises only Russian, punctuates by itself and is the fastest thing in the catalogue. In the program's own test it turned an eleven-second phrase into text in 0.47 seconds against 11.6 seconds for Whisper Medium, in 277 MB of memory instead of 814. GigaAM v2 is the previous version, just as fast but without punctuation. Parakeet TDT 0.6B v3 covers Russian as well and is a good second choice if you switch between several European languages. For words that appear while you speak, Nemotron 3.5 streams live text to the plate. Whisper Medium q5 remains the model for translation: when GigaAM serves Russian, the program keeps the best installed Whisper standing by, and the plate offers to hand a phrase to it. Qwen3 ASR is heavier and slower but the most accurate when memory allows.
German
For German the recommended model is Parakeet TDT 0.6B v3: it covers 25 European languages, punctuates by itself and turns a sentence into text in well under a second. Nemotron 3.5 shows words on the plate while you are still speaking and handles German among its 40 languages, so pick it if live text matters more than the last bit of accuracy. Canary 180M is the small option: English, German, Spanish and French in a model that runs on modest hardware, and it translates between those four languages without Whisper. Whisper Medium q5 is the general choice when you dictate in many languages or want translation into English; the program keeps the best installed Whisper standing by even when Parakeet serves German. Qwen3 ASR is the most accurate model in the catalogue and also the heaviest, for computers with memory to spare.
French
For French the program recommends Parakeet TDT 0.6B v3, a narrow model for 25 European languages that punctuates by itself and answers in a fraction of the time Whisper needs. If you want words on the plate while you are still speaking, take Nemotron 3.5, which streams live text and lists French among its 40 languages. Canary 180M is the light option for an older computer: English, German, Spanish and French in one small model, with translation between those four built in, so a French phrase can come out in English without Whisper. Whisper Medium q5 stays the general choice for many languages at once and for translation into the other interface languages; the program keeps the best installed Whisper standing by even when Parakeet serves French. Qwen3 ASR is the most accurate and the heaviest model in the catalogue, for a computer with memory to spare.
Spanish
For Spanish the recommended model is Parakeet TDT 0.6B v3: 25 European languages in a narrow model that punctuates by itself and turns a sentence into text in well under a second on a CPU. Nemotron 3.5 shows words on the plate while you speak and covers Spanish among its 40 languages; choose it when live text matters more than the last bit of accuracy. Canary 180M is the small option for an older computer: English, German, Spanish and French in one model, with translation between those four built in, so a Spanish phrase can come out in English without Whisper. Whisper Medium q5 remains the general choice for many languages and for translation into the other interface languages; the program keeps the best installed Whisper standing by even when Parakeet serves Spanish. Qwen3 ASR is the most accurate model in the catalogue and also the heaviest.
Italian
For Italian the program recommends Parakeet TDT 0.6B v3: 25 European languages in a narrow model that punctuates by itself and turns a sentence into text in well under a second on a CPU. Nemotron 3.5 shows words on the plate while you speak and covers Italian among its 40 languages; choose it when live text matters more than the last bit of accuracy. Whisper Medium q5 is the general model for many languages at once and for translation into English or another interface language, and the program keeps the best installed Whisper standing by even when Parakeet serves Italian. On a weak machine, Whisper Small or Base keep the program light at some cost in accuracy, and Canary 180M does not cover Italian. Qwen3 ASR is the most accurate model in the catalogue and also the heaviest, for a computer that has memory to spare.
Polish
For Polish the recommended model is Parakeet TDT 0.6B v3: a narrow model for 25 European languages that punctuates by itself and answers in a fraction of the time Whisper needs, which matters for a language with long words and rich endings. Nemotron 3.5 streams words to the plate while you are still speaking and lists Polish among its 40 languages; go back to Parakeet if a phrase comes out wrong. Whisper Medium q5 stays the general choice when you also want translation into English or another interface language, and the program keeps the best installed Whisper standing by even when Parakeet serves Polish. On a weak machine, Whisper Small or Base keep the program light at some cost in accuracy. Qwen3 ASR covers Polish as well and is the most accurate model in the catalogue, but also the heaviest.
Your own models
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.
How translation works
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.
A longer read on the three model families, with sizes and seconds: Whisper, Parakeet or Nemotron. And one page about dictation in your own language: English dictation on Windows, offline.