karaok-AI: exploring the limits in complex conditions

[EtienneAb3d]

In order to test and explore the speech-to-text limits in complex conditions, I wrote an automatic karaoke creator/editor/viewer.
Perfect for preparing fun DJ/karaoke parties in minutes!
https://github.com/EtienneAb3d/karaok-AI

Any advice/help would be appreciated on the way to improve the transcription quality and/or the timestamp labelling (of course, the interface is still a bit rough. It will be improved. This is not the real subject for a discussion here).
;-)