Closed DeboBurro closed 1 year ago
oh, I found why. All the input data lists u
, d
, e
, s
need to be (16000, 1)
in order to run inference.
Yup, all meta-af modules expect a final channels/microphones dimension.
think as long as the data size is larger than the window size or frame size or hop size.
According to the fig.4 in the Meta-AF: Meta-Learning for Adaptive Filters paper, I created a function that takes 2 lists,
u
andd
, as input and I want it to be able to get the prediction from the pretrained model. But when I run it, it shows the errorValueError: 'TimeChanCoupledGRU/~/linear/w' with retrieved shape (20, 32) does not match shape=[16, 32] dtype=dtype('complex64')
. Do you know why? Also, the prediction meansy
in the paper, right?I copied the
aec_eval.py
toaec_get_output.py
and created couple functions. But mostly the same.error log: