trying to export labels but theyre all saving the entire audio

export multiple

if i am to train a voice on spleeter'd audio, would using the default vocoder be better than finetuning one? i feel like it would because the default is trained on clean audio, while the voice audio i have is low-quality in part due to conversion from spectrogram (i think)

or i suppose finetune it on a similar voice that does have clean audio 🤔

...or just use the clean audio i do have that i've potentially underestimated the quantity of

can someone help me train a model with my own dataset (my voice) in romanian ?

i tried tensorflow tts but i see it doesent work in all languages by default and i have to do some things i dont know how

problem

I'm trying to make a TalkNet model, but these errors keep popping up. Is there any way to fix these errors?

did you follow the installment procedure in readme.md

also do you have exactly python 3.7 else there's gonna be problems

i installed 3.7 in a separate folder specifically for this thing

daily link to fixed notebook https://colab.research.google.com/drive/1KSeOCgyFXTYgVIvvPhOMVXlLvMXwiQMM?usp=sharing

i am curious as to how this error is even happening in the first place who is the person at pytorch who broke torchaudio

How do you install it in a different folder?

I tried that. The same errors kept popping up.

if you already have 3.7 then that's not necessary

just run the commands from readme.md to get all the dependencies

._.

🚎

didn't really preform well

might try again at a another time

Why is this happening?

Is this a bad thing? Also when it does more epoch's it sometimes changes to the diagonal incline. I'm just not sure what this is really.

Does somebody know why does my model sound like the default female voice instead of the actual model?

how many clips are you training on

and for how long

only 45 epochs in, and im so surprised!

im gonna say it is because my google drive code had the word pog in it as good luck lol

Dutch Van der Linde