Option to automatically create a karaoke/instrumental version by stripping vocals from audio

bohning / usdb_syncer

MIT License

67 stars 9 forks source link

Option to automatically create a karaoke/instrumental version by stripping vocals from audio #168

Open bohning opened 1 year ago

bohning commented 1 year ago

Spleeter or demucsV4 do a pretty good job in separating vocals and instrumental part and could be used for that.

Pro:

real karaoke without the original singer’s voice

Con:

will take long without GPU/CUDA
will bloat the executable

Baklap4 commented 1 year ago

Yes please <3

Would these be different text files? or an extra tag within the same txt file?

You could make it an optional buildflag and provide 2 binaries one with one without

Baklap4 commented 4 weeks ago

We can use spleeter to get the vocal track and use it with the #VOCALS and #INSTRUMENTAL to switch vocals on/off for a particular song. Thus no need for a seperate songfile

I know performous can already use this data but rather with #VOCALS and #AUDIO / #MP3 by using the shortcut ctrl + v

Spleeter can run locally, but we could also add in a option to talk with their api? https://github.com/deezer/spleeter/wiki/4.-API-Reference

bohning commented 4 weeks ago

There is already an instrumental_version branch that uses demucs. Not sure if demucs or spleeter is better suited for the audio track separating - I went for demucs as it was the recommendation by UltraSinger dev.

Baklap4 commented 4 weeks ago

Oewhh awesome! I'll have look there! Thanks :D