I'm not aware of a specific ml model for this, but I know there is some adjacent ones that pull just the lyrics. Perhaps someone needs to publicly release one to counter this type of behaviour.