Choosing a transcription engine
Applies to WhisprDesk 1.9.0 · Updated 2026-11-16
Choose local transcription and your voice never leaves your Mac. Prefer a cloud model? Bring your own key. Your choice. Both are set up in the same place, the Models tab in the WhisprDesk app, and you can switch at any time.
Before you start
Open the WhisprDesk app: click its icon in the Dock, or open it from your Applications folder. The WhisprDesk window opens on the Transcripts tab. If WhisprDesk is running but you don't see its window, click the WhisprDesk icon in the menu bar at the top right of your screen and choose Open WhisprDesk. The Models tab is the last tab across the top of the window.
Transcription models (Whisper, NVIDIA Parakeet, and cloud providers) are chosen in Models > Speech to text. AI text cleanup models are set up separately in Models > AI cleanup (see AI cleanup).
Local or cloud
| Local | Cloud | |
|---|---|---|
| Where your audio goes | Stays on your Mac | Sent directly from your Mac to the provider you choose |
| What you need | One model download | An API key from that provider |
| Cost | Included | Billed by the provider to your own account |
| Works offline | Yes, once the model is downloaded | No |
WhisprDesk does not run its own transcription servers. In cloud mode, your audio goes straight to the provider whose key you entered, under that provider's terms.
The app starts in Cloud mode until you choose a local model. Nothing is transcribed on your Mac until you download one and click Use this model.
Open the Models tab
Where: WhisprDesk > Models > Speech to text
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the Models tab at the top of the window. Three buttons appear at the top: Speech to text, AI cleanup, and Agent. Click Speech to text. The page is called Model Library. WhisprDesk remembers which of the three you opened last.
- At the top, a banner shows Now transcribing with and your active model. Below it are two tabs: Cloud (your key) and On this Mac.
How choosing a model works
On the Speech to text page, one rule applies: a model becomes active only when you click Use this model.
- Downloading a model never switches to it.
- Clicking between the Cloud (your key) and On this Mac tabs, or between provider tabs, only lets you look. It does not change what you are using.
- The active model shows In use. Other downloaded local models show Downloaded and a Use this model button.
- A cloud model can only be used after you enter that provider's key.
Set up local transcription with NVIDIA Parakeet
Where: WhisprDesk > Models > Speech to text > On this Mac > NVIDIA Parakeet
NVIDIA Parakeet is a simple local choice: one download of about 680 MB, it detects the language automatically, and it supports 25 languages. If you rely on your Vocabulary for names and special terms, note that Parakeet does not use it; a local Whisper model does.
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the Models tab at the top of the window, then click Speech to text.
- Click the On this Mac tab.
- Click the NVIDIA Parakeet tab.
- Find Parakeet TDT 0.6B and click Download. The download is about 680 MB and can take a few minutes depending on your internet connection. You can click Cancel while it downloads.
- When the download finishes, the model shows Downloaded. Click Use this model.
- The model now shows In use, and the banner at the top shows it with on this Mac. Your next dictation uses it.
Parakeet supports: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish, and Ukrainian. For other languages, use a Whisper model.
Set up local transcription with Whisper
Where: WhisprDesk > Models > Speech to text > On this Mac > OpenAI Whisper
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the Models tab at the top of the window, then click Speech to text.
- Click the On this Mac tab.
- Click the OpenAI Whisper tab.
- Pick a size and click Download.
- When it finishes, click Use this model. It shows In use.
| Model | Download size |
|---|---|
| Tiny | 75 MB |
| Base | 142 MB |
| Small | 466 MB |
| Medium | 1.5 GB |
| Large | 3 GB |
| Turbo | 1.6 GB |
Whisper models download from Hugging Face, and Parakeet from the sherpa-onnx project on GitHub. After that, local transcription needs no internet connection.
Set up cloud transcription
Where: WhisprDesk > Models > Speech to text > Cloud (your key)
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the Models tab at the top of the window, then click Speech to text.
- Click the Cloud (your key) tab.
- Click a provider tab: OpenAI, Groq, Mistral, or Custom.
- Paste your key into API Key. No key yet? Click Get key. It opens the provider's website, where you sign in and create one.
- Click a model. It shows Selected.
- Click Use this model. The model shows In use, and the banner shows the model with via and the provider's name.
Full details: Cloud API keys.
Check which model is active
Where: WhisprDesk > Models > Speech to text (banner at the top), or the Transcripts tab
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the Models tab at the top of the window, then click Speech to text.
- Look at the banner at the top. Under Now transcribing with it shows the model and either on this Mac or via your provider.
- If it shows No model selected instead, dictation will not work until you download a local model or enter a cloud key, and then click Use this model.
The Transcripts tab also shows the active model under Transcribing with. Click it to open Models > Speech to text.
Delete a model
Where: WhisprDesk > Models > Speech to text > On this Mac
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the Models tab, click Speech to text, then click the On this Mac tab.
- Click the OpenAI Whisper or NVIDIA Parakeet tab.
- Click the trash icon next to the downloaded model.
- Confirm Delete Model. You can download it again later.
If you delete the model you are using, no model is selected until you choose another one and click Use this model.
To remove every downloaded model at once:
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the gear button at the top right of the window to open Settings, then click System.
- Find Model cache and click Clear Cache. This removes Whisper, Parakeet, and local AI cleanup models.
Set your transcription language
Where: WhisprDesk > Settings > Preferences > Language
- Open the WhisprDesk app: click its icon in the Dock (or open it from your Applications folder).
- Click the gear button at the top right of the window to open Settings, then click Preferences.
- Under Language, set Transcription language to the language you speak. Auto-detect is the default.
The list has 58 languages. Parakeet supports the 25 listed above. With a cloud provider, supported languages depend on that provider.
Tip: if the wrong language or alphabet appears, pick your language instead of Auto-detect. Some models confuse languages that sound alike (for example Urdu and Hindi) or write them in the wrong alphabet. When you choose a language here, WhisprDesk sends it to the model as a hint: with Whisper on this Mac, and with OpenAI, Groq, Mistral, and most custom providers in the cloud. It can make a big difference, though results still depend on the model. NVIDIA Parakeet detects the language on its own and only covers its 25 languages, so for other languages use a Whisper model or a cloud model.
Very short recordings (under 4 seconds) that come back in a different alphabet from your language are treated as background noise and skipped. If you dictate short phrases in another language, set that language here.