Dev pr3 : small changes for chinese implementation

[kirianguiller]

Hi guys,

Here the third split of the PR I did earlier this week ! (finally the last one :) )

In this PR,

• I've changed the name of the phoneme language variable ( 'chinese-mandarin' -> 'zh-CN') for more conciseness.
• I've changed the .models.json google drive identifiers for a new model that trained 130k epochs.

Thanks for your reviews !

[erogol]

This PR still includes abstraction classes and other changes than only the Chinese model.

Is it international or you forgot to remove the other changes?

[kirianguiller]

I was thinking that if you merged the pr1 first, then I can rebase pr2 and pr3 (so the commits in the pr1 are added to the dev and then they are not being compared anymore for pr2 and pr3).

Then, if doing the same after merging pr2, i can rebase pr3 with the new dev, and the only differents commits between this new dev and the pr3 will be the commits regarding chinese implementation.

I was not sure of doing these PRs, and I found some people online recommending this way. You want to do it another war ?

[kirianguiller]

Hey, I've just changed the branch for only including the changes relative to the new chinese model

[erogol]

THIS IS AWESOME !! THX !!

[erogol]

@kirianguiller btw what vocoder does this model use ?

[kirianguiller]

The following melgan-multiband :

"ljspeech":{
                "multiband-melgan":{
                    "model_file": "1Ty5DZdOc0F7OTGj9oJThYbL5iVu_2G0K",
                    "config_file": "1Rd0R_nRCrbjEdpOwq6XwZAktvugiBvmu",
                    "stats_file": "11oY3Tv0kQtxK_JPgxrfesa99maVXHNxU",
                    "commit": "ea976b0",
                    "author": "Eren Gölge @erogol",
                    "license": "MPL",
                    "contact":"erengolge@gmail.com"
                }
            }

It's working really well :)