>>86359075UVR5:
one pass of mdx23c-instvoc hq2. this splits the audio into vocals(with some licking) and intstrumentals(with the rest of the licking).
the reason it has licking in both is because it is trained to find music and singing, not gooner asmr.
i then took the instrumentals and ran two passes of demucs. it splits audio into vocals and different intstruments. i forgot which models.
both times it placed the licking in the drum track. that allowed me to isolate the licking by cutting away the drums and then combine it
with the vocals. some got lost in the process but i think i got most of it.