because talknet

i dont think you need to do anything with talknet

i dont think talknet does validation exactly like tacotron2

whats the cause of the blocking of that word

because its stooopid

ask mr bean he made the bot

not exactly

validation can go up but it can still decrease overall (see: tensorboard screenshot above)

you can run tensorboard on colab but i havent had a good experience with it since it would never open in a separate tab

i dont like using tensorboard from within a cells output in my case i run tensorboard while training (locally) and it updates somewhat in real time

i guess when the cell stops

i normally stick with 250 with epochs

keeping it on for too long makes the voice robotic

for some reason...

(uberduck uses sortanons fork)

when it doesn't improve for a few iterations in a row

i think they implement the models as pytorch modules though so i am thinking about something

e.g. my children's song dataset model bottomed out at 0.25

well how much did it improve since those hours ago

tacotron HATES 29.wav

yeah

i think my model forgot how to press the stop button

it cut out

I'm trying to update a model but I need help. Could one of you do me a favor and fix my voice clips for me because the mel spectograms ain't working?

is there any ways to mount drives like the old way in google colab

found it

drive._mount('/content/drive')

total noob question. I have a 2080ti with 11gb of memory. Can that do any damage with talknet?

i mean the k80 (the free colab gpu) has 12gb per core so it should be good on at least the vram front as for performance, i don't think gaming gpu·s are often benchmarked with the same things that machine learning gpu·s are so it'll be hard to get exact numbers, but it should definitely function at the very least

Thanks. That was encouraging enough for me to try and do the tutorial