Understanding feat tensor dimensions

[mrgloom]

What does feat tensor dimensions mean?
https://github.com/facebookresearch/loop/blob/331cbd0ac2c5824998424095f91b80affff50d86/generate.py#L124

Looks like when generation is done from text feat tensor empty before inference and after, so it's just a way of input of precomputed features when loading from npz?

[mrgloom]

63 is number of features for vocoder according to paper: https://arxiv.org/pdf/1707.06588.pdf