I am evaluating the testset with 24K sampling rate as the readme instructs. However, I cannot reproduce the evaluation result in the paper: if I use LSD with log10(as the paper states), then I obtain 0.36 LSD which is much lower than the LSD reported in the paper(0.69). If I use LSD with ln(as some other papers suggest), then I obtain 0.82 LSD which is higher than the LSD reported in the paper.