got it

loading state dict?

dont use finalized

use latest checkpoint

is the python3 error something to worry about?

yes, something went wrong in setup and setup got ran twice

kill the runtime and rerun

created this

did step 3 run correctly

everything shouldve ran correctly beforehand and the files it says dont have any directory appear in files

if its trying to assign output to

{relative_path}

without it being defined then something must have gone awry in step 3

i might try doing the whole notebook over in the morning (for me) it should clear out problems that happened beforehand

Am now getting an error

Nvm

The training session automatically timed out at 1200 cycles. Is there a way to continue it?

fixed the python issue when's a good time to stop the training?

I just let it train until it disconnects me tbh

Keep in mind that if you do that, your dataset needs to be in drive so it outputs the hifigan model to there

any reason for this?

I think that's an old notebook Use one of the ones in #841437191073955920

what cell is this on?

for copying the hifigen and google drive

its on the uberduck pipeline interferencing

using the cpu tacotron2 hifigen synthesis thats in the notebooks channel, it should work, but i keep getting this

id isnt reading correctly

either the file isnt anyone with link can read or the id was put in wrong

got it, thanks :)

How do multispeakers work and what are their applications?

Multispeaker banks work by adding a speaker id to the end of the line of the transcription. The speaker ids separate the different voices. The current pipeline notebook does this automatically as you load different datasets for your multiple speakers.

There are a few applications for multispeaker models. One is for training a small dataset that is too small to even work on aitch or crust. If you train a small dataset alongside one or more bigger ones, it will be able to learn to speak. Another application is multilanguage training. You can train a model to speak in more than one language by training different language datasets, or you can convert a dataset from its source language to a new language by transcribing it accordingly and training it alongside a complete dataset of the target language.