Open zeehio opened 7 years ago
In all versions of the voice I have, the filters coefficients are as given in the attached files. Interestingly, for the first h_0 - h_2 they are the same, but for h_3 and h_4 they differ.
But to be honest, I'm currently training a few versions with different MEF orders and I didn't notice any perceivable differences yet. Currently it seems to me that the main benefit is to get rid of the overflow.
I am using a zero for now. See https://github.com/MycroftAI/mimic/issues/132