Contribution in a nutshell

So far we have been conducting everything on-the-fly when doing evaluation. However, there might be some cases where the users would like to use embeddings for some other purposes (and it is the trend nowadays for speech processing in my opinion). Therefore, it would be helpful if we can provide a simple script that extract the embeddings using our models, given a list of utterances and their wav paths.

This PR also cleans one line in the hparams file on data folder.

Notes for reviewing (optional)

This change has these implication which might need attention over here; —how should we tackle this?

Pre-review

• (if applicable) add minor comments on the code
• Perform testing by somebody else