seq

[whh07141]

hi，I want know how to set the T'_i ,I have extract speech representation

[Lhx94As]

Hi,

You can have a pre-processing before or during training (torch reshape), by reshaping the features from (T, feat_dim) into the shape like (T/20, 20, feat_dim) and discarding the left frames (e.g., (101, 20) => (5, 20, 20) )

Hope this can help.