r4l v0.0.4
Quality of Life & Memory Awareness
tl;dr This is a small update, I'm still working on #9 and #13, which will be fixed by 0.0.5 and 0.1.0, respectively. Otherwise, mostly quality of life adjustments and small fixes to minor bugs.
Memory
- I added a few lines to decrease memory usage by dropping the wav file in RAM to compressed mp3 whenever this process is using more than 60% of available ram (this will be configurable in cl/ui later, is configurable in api)
QOL
- The decoder steps are now dependent on the sentence being read, which decreases the number of cursed tangents due to short sentences.
- Fixed sentence splitting issues
- Added some incremental changes to solve header/footer issues (openCV solution next version)
Stress tests
I unintentionally ran this for 50+ hours on texts and it survived the affair, not even crashing out! I'm very proud of it :)
pypi link