Languages and speech recognition models
Choose a dictation language, compare supported speech models, and follow multilingual recording guidance.
Choose a dictation language
Open Home, select Dictation Language, and choose the language you plan to speak. Whisperstream selects a supported speech-recognition model and, if needed, asks before downloading it. Wait for the download to finish before dictating.
The active model appears under Voice Recognition Model on Home. For a new selection, Whisperstream uses:
- NVIDIA Parakeet v2 for English.
- NVIDIA Parakeet v3 for supported non-English languages.
- Qwen3 ASR when Parakeet does not support the selected language.
If the active multilingual model already supports the new language, Whisperstream may keep that model instead of downloading another one. When you switch to English from a multilingual model, the app may let you keep the current model or switch to English-only Parakeet.
Important: Keep each recording to one language. By default, multilingual models detect the language separately for each recording, but changing languages mid-recording can reduce accuracy.
Compare speech-recognition models
| Model | Used for | Languages | Approximate download |
|---|---|---|---|
| NVIDIA Parakeet v2 | Default English recognition | English | 661 MB |
| NVIDIA Parakeet v3 | Default recognition for supported non-English languages | 25 languages | 670 MB |
| Qwen3 ASR | Recognition for languages outside Parakeet's range | 30 languages | 987 MB |
After the selected model is downloaded, speech recognition runs on your PC without an internet connection. Selecting a language that needs another model prompts you to download it.
These models produce the raw transcript. For optional cleanup after recognition, see Enhancement providers.
Supported languages
Whisperstream offers 39 dictation languages. Default model shows what a new language selection uses. The current multilingual model may remain active when it already supports the selected language.
| Language | Supported models | Default model |
|---|---|---|
| Arabic | Qwen3 ASR | Qwen3 ASR |
| Bulgarian | Parakeet v3 | Parakeet v3 |
| Cantonese | Qwen3 ASR | Qwen3 ASR |
| Chinese | Qwen3 ASR | Qwen3 ASR |
| Croatian | Parakeet v3 | Parakeet v3 |
| Czech | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Danish | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Dutch | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| English | Parakeet v2, Parakeet v3, Qwen3 ASR | Parakeet v2 |
| Estonian | Parakeet v3 | Parakeet v3 |
| Filipino | Qwen3 ASR | Qwen3 ASR |
| Finnish | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| French | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| German | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Greek | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Hindi | Qwen3 ASR | Qwen3 ASR |
| Hungarian | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Indonesian | Qwen3 ASR | Qwen3 ASR |
| Italian | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Japanese | Qwen3 ASR | Qwen3 ASR |
| Korean | Qwen3 ASR | Qwen3 ASR |
| Latvian | Parakeet v3 | Parakeet v3 |
| Lithuanian | Parakeet v3 | Parakeet v3 |
| Macedonian | Qwen3 ASR | Qwen3 ASR |
| Malay | Qwen3 ASR | Qwen3 ASR |
| Maltese | Parakeet v3 | Parakeet v3 |
| Persian | Qwen3 ASR | Qwen3 ASR |
| Polish | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Portuguese | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Romanian | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Russian | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Slovak | Parakeet v3 | Parakeet v3 |
| Slovenian | Parakeet v3 | Parakeet v3 |
| Spanish | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Swedish | Parakeet v3, Qwen3 ASR | Parakeet v3 |
| Thai | Qwen3 ASR | Qwen3 ASR |
| Turkish | Qwen3 ASR | Qwen3 ASR |
| Ukrainian | Parakeet v3 | Parakeet v3 |
| Vietnamese | Qwen3 ASR | Qwen3 ASR |
If recognition chooses the wrong language
- Confirm that Dictation Language matches the language you are speaking.
- Use one language for the entire recording.
- If Qwen3 ASR repeatedly detects the wrong language, open Home > Advanced and turn on Force selected language.
Force selected language is available only while Qwen3 ASR is active. Turn it off when you want Qwen to detect the language automatically for each recording.