btw when should I stop pepper?

yo @User what do I do

i don't really know

i haven't trained that many models

but let's say as many epochs as you have clips, up to 250

advantage of my dataset loader: one can give a prepared notebook to a kid who wants a model trained and tell them to

run all

and the google drive login thing and they can train

heck maybe even merge all the cells into one giant one for maximum foolproofing

in other news my learning rate is now extremely low due to the whole epoch thing, took it an hour to go down from .147 to .137

-or no wait that's 25 epochs so it's not that bad

-no wait that's not because epochs, since it calculates the learning rate based on iterations afaict (which aren't affected by dataset size)

still i might wanna increase it to help get rid of the learned noise

How do I add a character in the inference notebook so I can use my hifi-gan finetuned pepper model

1) double-click the cell to reveal the code b) add a name to the list that's right under the comments, to the right of

character = "whatever" #@param

iii) use the name to attach to the big ol' chain of `elif`s under that

do I format like this elif character == "Character (fine-tuned)": TACOTRON2_ID = "ID" HIFIGAN_ID = "ID"

yeah

OK

i'd imagine it'd be super easy to add a text field for custom hifigans

So I include the tacotron2 id and hifigan id

got it

yeah

comically large if elif

ppp's ngrok notebook uses a list instead and checks which entry to use with

.index()

, calling it a good way to avoid elifs

and yeah it is

though i'd use a dict instead

Did it

and it worked 😎

spoke to soon

oh yeah i love dicts

NameError Traceback (most recent call last) in () 87 try: ---> 88 initialized 89 except NameError: NameError: name 'initialized' is not defined During handling of the above exception, another exception occurred: ModuleNotFoundError Traceback (most recent call last) in () 118 import torch 119 import json --> 120 from hparams import create_hparams 121 from model import Tacotron2 122 from layers import TacotronSTFT ModuleNotFoundError: No module named 'hparams' --------------------------------------------------------------------------- NOTE: If your import is failing due to a missing package, you can manually install dependencies using either !pip or !apt. To view examples of installing some common dependencies, click the "Open Examples" button below. ---------------------------------------------------------------------------

in fact the script to check for new voices in uber duck dot ai that i made uses dicts