sampling rate for JDC network for use in HiFTNet

Hi,

If I want to train the F0 model from scratch in the other repo, do I need to train the JDC with a 22k sampling rate in order to use it for HiFTNet, or is it okay to use the 24k sampling rate (using the default SPECT and MEL extraction parameters in meldataset.py) for training the JDC network?

Thanks
