For full disclosure this process uses AI but it does not use any kind of AI generation and isn’t possible to this degree without it.
This tutorial uses 1 tool with an extra plugin.
Ultimate Voice Remover - A free opensource AI audio separating program.
Cinematic Sound Demixing - The model used to separate the music and the sound effects
Install UVR (Ultimate Voice Remover) and download CSD (Cinematic Sound Demixing).
Navigate to where you installed UVR and open “models\Demucs_Models\v3_v4_repo”. From here you’ll likely see two files. For me, these files are “955717e8-8726e21a.th” and “htdemucs.yaml”
Screenshot of the folder
Copy a single .th file into this directory. For this example I’ve used “97d170e1-a778de4a.th” and then copy “htdemucs.yaml”, paste the copy in this directory and rename it. I’ve renamed mine to “Cinematic_Sound_Demixing.yaml”
Screenshot of added files
Now open “Cinematic_Sound_Demixing.yaml” - to open it, I use notepad++. It should read “models: [‘955717e8’]” change this to “models: [‘97d170e1’]”. Now we’re ready to open UVR.
Screenshot of Edited File
Open UVR and use Select Input to chose the track you wish to separate the SFX and music. Use Select Output to specify the outfit of the separated audio. Choose Process Model “Demucs” and then Demucs Model “Cinematic_Sound_Demixing” (This will be whatever you named the .yaml file). Once that’s all selected, make sure the other settings are to your liking and press “Start Processing”.
If you receive and error here, don’t be alarmed. Just check your Output folder.
In your output folder you should have a (Bass), (Drums) and (Other) version of your original file. From what I can tell, Bass is the separated SFX, Drums are the music and Other is voice.
From here, you can import your sound effects and new audio for the scene you’re rescoring!
Some things to keep in mind
This works best with stereo audio, so if you’re editing 5.1 like I do, just export out 2 channels at a time and patch them back together before importing them back into your project.
