Clearing out background noise after using a stem splitter

[imkira3]

So recently I took to dubbing the missing episodes of the DragonBall Z lost dub. If you didn't know, the show did have a second english version that is currently stuck in a sort of legal limbo. This show was never released on VHS, DVD, or any sort of medium, but fans can be tenacious. There are people out there who went around finding every VHS recording of the show they could, and so the last episode of the lost dub was finally found a few years ago, and there are new technologies out there I can pair with my own creations to take voice clips of the characters in the show and re-dub the missing episodes, because even though all existing episodes are found, there is a gap in the show that was never dubbed by Westwood or Saban. The process is complex, I won't go into details, but needless to say I need clear voice samples for any of this to work. And so I used a stem splitter to split the vocals and the music of the show into two clips, but the process isn't perfect, and I'm having trouble imagining how I might clear up the last of the musical interference, or I might have to throw out alot of my voice samples. Note: This is technically classified as a nonprofit fan-based parody, like what TeamFourStar did. I share no affiliation with Akira Toriyama.

[davexnet]

Try some of the other stem splitters on the 'net and/or post a sample or two here, perhaps other members will have some idea's ....

[imkira3]

Sure, heres a sample. As for using a better stem splitter, I already used the best rated stem splitter, LALAL.AI. Sample: mediafire.com/file/1y04nohnt4hy1xa/53C.mp3

[imkira3]

I made a bit of progress, I'm gonna subject all the files to a second conversion using LALAL. The first conversion was to split voice and music, the second will be voice and noise which seems to have a positive effect. After that I'll try some form of echo reduction tech, I would love to hear some of your favourite echo reduction methods, that could help. It would probably need to be ai-based since the amount of echo varies greatly

[imkira3]

Heres another sample, this ones after the voice/noise and voice/music split: https://www.mediafire.com/file/2plec0ufp4f4org/1B.mp3

[imkira3]

Here's the same track after being split by vocali.se, as it turns out a second pass with vocali.se is better than a second pass with LALAL.AI, seems their algorithms work well together: https://www.mediafire.com/file/pc3pakyrhzthjv7/1B_%25282%2529.mp3

[davexnet]

Here's the same track after being split by vocali.se, as it turns out a second pass with vocali.se is better than a second pass with LALAL.AI, seems their algorithms work well together: https://www.mediafire.com/file/pc3pakyrhzthjv7/1B_%25282%2529.mp3

Yeah I thought this one was pretty clean. I tried Izotope RX dialogue isolate but it was not as good,
looks like Izotope have some work to do on their separations

[imkira3]

Here's the same track after being split by vocali.se, as it turns out a second pass with vocali.se is better than a second pass with LALAL.AI, seems their algorithms work well together: https://www.mediafire.com/file/pc3pakyrhzthjv7/1B_%25282%2529.mp3

Yeah I thought this one was pretty clean. I tried Izotope RX dialogue isolate but it was not as good,
looks like Izotope have some work to do on their separations

Okay I'll scratch that one off the list, I'm trying all the best stem splitters