Improved detection and segmentation - dataset balancing

Improvements to automation were made, including a much better automatic threshold detection, automatic background noise collection and automatic dataset balancing. I have also re-added manual thresholds so that you can tweak the automatic results yourself if you aren't satisfied. Some performance improvements were made as well, which should improve training time drastically.
Some crashes were fixed as well.

Auto-recording

Automatically decides on thresholds for your recording sessions
Changes recording data from small wav files to SRT files

Improved recording

Improved the recording menu by allowing multiple microphones to record at the same time
Also improved resource usage during recording