Comment by htrp

subbed audio training data (much better than cc data) is better