reproduction steps for inference

[ghost]

are all the required preprocessing steps:

1. acquire dataset, checkpoint, source
2. train spm
3. hubert feature extraction
4. run fairseq

or are there any missing parts?

[ghost]

Also, it's a bit opaque what we're meant to do with hubert, precisely

[Ajyy]

Could you give more details about inference, such as what tasks you are doing, what do you want to do?