what's a gta synthesis

Step 10: Generate GTA spectrograms. This will help HiFi-GAN learn what your TalkNet model sounds like. If this step fails, make sure you've finished training the spectrogram generator.

They're basically just the model outputting sound right outta the tts pipeline

And the vocoder smoothes out the spectogram noise and applies a filter to it to make it sounds better

i know that much

unfortunately i closed the training notebook because i didn't know i would need it

i still have the files though

Noice, do you have those raw spectogram from before they were put in the vocoder?

no i mean like

the files that it keeps at the end of the whole process

this

What's in TallNetSpecs?

Hmmm

And I don't suppose the checkpoint had any example generated files?

nnnope

Dang

O well

i have started a new model with 253 audio clips and started training it but almost all the time the graph looks like this and its 125 epochs in out of 250. Is this training good and should i leave it to continue or?

nope

it's an actual line that spy has

Kill it with fire!

Thats great news to hear when i've spent two hours training it already lol

you want me to try again?

I mean I used less than 4.5 mins of audio for Queen elizabeth

But the Queen speaks very slow and concisely so that probably very much helps.

Hmmm

Do you have that dataset?

Try rap god lyrics

Wavs https://drive.google.com/file/d/1yStk02VZmAH-T_uIef7pLZQHWD78DtLW/view?usp=sharing, List https://drive.google.com/file/d/1RzFFexdhuCrHxv0oftAwFf74BRkt8jnR/view?usp=sharing