Output generation from local model takes long after inactivity

[stefan-girlich]

Describe the bug
After a few minutes of inactivity, the next speech-to-text operation takes up to 20 seconds to show a result (end of speech input to text pasted).

To Reproduce
Steps to reproduce the behavior:

1. Set up local model (i.e. Whisper large)
2. Use OpenWhispr as usual; outputs are pasted almost immediately.
3. Do not use OpenWhiser for a while (estimation: 10-15min)
4. On the next invokation, OpenWhispr will take around 20s to produce a result, even for very short inputs

Expected behavior
STT outputs should be pasted immediately after inactivity.

I may be mistaken, but I think this problem only started occurring a week ago. I could imagine that a procedure for unloading/reloading the model has been added.

Desktop

• MacOS Tahoe 26.3, Apple M2 Max
• OpenWhispr 1.6.10