0.44.0

• Narration on an NVIDIA graphics card. Settings → Narration → GPU
  acceleration appears where the driver reports a card CUDA 12 can use, and
  downloads what the card needs: about 2 GB, 2.6 GB once unpacked, every file
  checked against its published checksum. Switched on, the neural voices run on
  the card: on a GTX 1060, three times faster than four processor cores, with
  the whole app at 2% of the machine while it reads aloud. If the card or its
  driver turns the model down, narration uses the processor as before and
  Settings says why. The pack can be removed from the same place. On Linux the
  same pack is offered but has not yet been tried on a machine.