• DNN-based (deep neural network) vocal splitter working on Linux/Windows (with GPU acceleration) and Mac OSX (CPU only, because the stupid MacBook all use AMD GPU rather than Nvidia GPU, I am pretty sure Steve Jobs will not make such a decision)
• added essential Karaoke-system features: vocal/nonvocal play, pitch shift no need to start from beginning, audio delay, seek to play position, non-English text display, sort play queue by dragging, download video from non-Youtube websites (all video sites supported by yt-dlp)
• VLC directly renders into PyGame window, this dramatically reduces latency and totally eliminated transient screen glitch artifact when relaunching VLC for pitch shifting and vocal/nonvocal switch (this does not work in Mac OSX because the stupid MacOS does not allow one app to draw into a window created by another app for security reasons)
• in Linux, can capture screen (with audio) and stream to HTTP; in Windows, we can use WIFI Display, in Mac OSX, we can use AirPlay
• can toggle between full-screen and windowed mode; in windowed mode, the window is resizable