it sounded like this i need help with this

For me it sounded fucked up like that when I tried to train something where there was a problem with the transcript. and/or when I tried to train something but the wavs were too long. I was too lazy to re-do them. But my dataset with shorter wavs didn't sound like that. But that was also when i was synthesizing them in the notebook linked in the pipeline notebook. The FakeYou notebook refuses to work for me, things could be different there.

Also even with plenty of training and everything else being correct, my outputs still sounded weird and warbling, but doing a hifigan fixed it

There's one problem i've encountered while contributing to this site (since 6 months ago).

Training with an .npy dataset (a list.txt with .npy)

no longer works on some notebooks

and yet it will mindlessly change it to .npy

it might not happen on the legacy notebook

but it happen on the fakeyou training one

I know that this problem can be fixed by deleting the 0-1 second wavs

and the 10+ wavs

but a model would actually learn from the voice clips

and then it becomes good training

and then it would just straight up fall off

**harder than a toxic fandom**

that's what happened when i was training dinoroki's dataset

he probably didn't know about the 2-10 length range or smth

(not to be rude)

why does this happen?: fatal: destination path 'hifi-gan' already exists and is not an empty directory. --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) <ipython-input-6-f7cf7f61b692> in <cell line: 15>() 13 14 use_gpu = get_ipython().getoutput('nvidia-smi') ---> 15 use_gpu = not use_gpu.startswith("NVIDIA-SMI has failed") 16 17 #@markdown **To enable GPU for faster synthesis, head into Runtime -> Change Runtime Type and set Hardware accelerator to GPU.** AttributeError: 'SList' object has no attribute 'startswith'

summon <@327116758714417153>

oops

refresh and run again

If I in the transcipt of the model I type in something like wavs/#.wav|Nifty! (one-liner sentence) then the final voice models pitch would be significantly different from the original! What should I do?

It causes the voice to pitch down or up, either sounding like a monster, or a chipmunk

A one-liner in the transcript

Esp. if there are various

Oh no I just wasted my usage limits in Gogle colab

I figure it was due to me over-testing my model in the synthesis notebook

<@563902837931638784>