EXTRACTING SONGS FROM SOUP OPERA AUDIO MIX

[craneoprod-bit]

"Hi everyone. I'm doing audio post-production for a TV show and I've hit a wall. I need to isolate spoken dialogue from a background song (sung by a female vocalist). I'm currently using UVR v5.6.0 on a Mac. I've tried the Ensemble Mode (Min Spec) crossing Demucs and MDX-Net (Kim Vocal 2), but the sung vocals are bleeding into the dialogue frequencies. Does anyone have a specific Roformer or MDX configuration/preset that aggressively targets sung notes while preserving spoken consonants? Thanks!"