The speech model download seems stuck
Quick fix
You can keep dictating. VybeScript bundles the Whisper engine, which covers dictation until Parakeet, the default speech model, is ready. The main download is about 600 MB and happens once. Leave the app open on a working internet connection and it finishes by itself.
Step by step
- Open Settings and find the Transcription card. The Model row shows Downloading model… with a percentage while it works. It may then show Optimizing model for this Mac…. In the main window, the Local Engine row under System Readiness says Parakeet (preparing) during this time.
- Check your internet connection. The download needs one the first time. After that, dictation works offline.
- Check your free disk space. The download is about 600 MB. If you already use Parakeet and an upgrade is arriving in the background, VybeScript skips it when your Mac has less than about 3 GB free and tries again on a later launch. Free up some space and relaunch the app.
- If the Local Engine row says Parakeet unavailable and an orange message appears under the Speech engine buttons, click Retry next to it.
- Still waiting? Quit VybeScript from the Vybe menu in your menu bar, open it again, and let it run for a few minutes.
- If you need a result right now, choose Whisper under Speech engine. It works without the download, but it is slower.
What gets downloaded
- On a fresh install: the default speech model (Parakeet Ultra), about 600 MB, once, at first launch.
- The first time you use taught words with Parakeet: a small extra model that listens for your words, about 100 MB, once.
- If you used an earlier version of VybeScript: you keep dictating with the model you already have while the new one downloads in the background. That needs about 3 GB of free space. In newer versions, VybeScript removes the old Parakeet v3 model by itself once the new model has worked for a while, and tells you how much disk space that freed. Until then both models are on your Mac.
None of these downloads are repeated when VybeScript updates. The models are stored on your Mac outside the app, and on launch VybeScript downloads only what is missing.
Still not working?
The speech model is downloaded from Hugging Face (huggingface.co). If the percentage does not move for a long time, a network that blocks that site, such as some VPNs, firewalls or work and school filters, could be the reason. Try another network. If it still fails, send us a report: see Send a diagnostics report.