Documentation / Getting started
Choosing a speech model
The speech model turns your voice into text on your own computer. A bigger model is more accurate and slower.
The three models
- Fast (Whisper base, 148 MB): the fastest, with more mistakes, especially in Dutch.
- Balanced (Whisper small, 488 MB): fast and accurate. This is the one SayFeather picks for you.
- Accurate (Whisper large-v3-turbo, 574 MB): the most accurate, but slower.
Downloading and switching
The setup downloads the model once. After that, speech recognition needs no internet. To get another model, go to Speech model in Settings, choose Download next to it and then Use. The one in use is marked In use.
A bigger model is not always better for you: if your computer is slow, the Fast model keeps dictation snappy. See Slow dictation.
On Android
The Android app has the same three, also named Fast, Balanced and Accurate. It starts with Fast, because phones are slower than computers: a larger model hears names and fast speech better but takes noticeably longer. You can switch at any time.
Stuck? Email [email protected].