Crowd and dialogue separator

Crowd and dialogue separator

A live recording contains more than the performance: applause, audience chatter, cheers and whistles can dominate the spaces between phrases. This mode returns two files, Crowd / dialogue and Other. Its MVSEP origin is a crowd-removal model, so the useful starting question is how much audience sound can be separated from the performance. It does not deliver independent speech, music and sound-effects tracks. Use the target to inspect where audience activity rises, or audition Other when preparing a quieter live-music excerpt. Choose a section with both music and a transition into applause. That comparison reveals whether the performance stays natural as the crowd becomes louder. Dialogue in the label should not be read as speaker identification or a guarantee that every spoken word will move to the crowd layer. Check handclaps that coincide with snare hits, whistles near sung notes and spoken introductions over sustained music. These overlaps are harder to judge than applause during silence. Some audience presence may be worth keeping for a believable live atmosphere. Compare a moderate reduction with complete removal, and listen for changes in the room sound before making the final edit.