Yes but the link may be outdated

Oh That Might Be Why Ok

Are you using the one @Gosmokeless28 sent?

Did you run the setup cell?

yes

Try manually selecting the model using the "manual_select" function

Can Someone Remove The Music And SFX From This, I Don't Feel Like It https://drive.google.com/file/d/1S04uHEMk5j6WdP1pid9OxwdVs4Iz9kaG/view?usp=share_link

I Am Doing Phineas' Season 1 Voice

THAT'S A WHOLE AHH PHINEAS AND FERB EPISODE HOW IS THAT STILL UP💀

btw i'm doing it in the background

I Thought You Were Removing The SFX And Music For This

Am I correct in assuming that there should be no commas in transcript?

I can't get it to be under 31 mb without it being .wav tho

So why not convert it to a .wav file?

No

it becomes 31 mb all the time

Does it make no difference how to write numbers in a transcript, in letters or in digits?

Writing numbers in word form is the best way to transcribe them

Got it

i did it, it's not wav tho https://cdn.discordapp.com/attachments/994486394049282058/1104904105236893796/8mb.video-JzR-UAWl3zK8.m4a

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.00 MiB (GPU 0; 14.75 GiB total capacity; 12.67 GiB already allocated; 832.00 KiB free; 13.61 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

What Site Did You Use?

is there a max amplitude to use to avoid auto clipping and the audio amplitude out of range warning on the files?

NameError Traceback (most recent call last) in () 6 #@markdown #### Import the transcript .txt file into the /content/tacotron2/filelists folder on the left. 7 Training_file = "filelists/transcription.txt" #@param {type: "string"} ----> 8 hparams.training_files = Training_file 9 hparams.validation_files = Training_file 10 NameError: name 'hparams' is not defined

Is that even a problem? That's just in regard to the TensorBoard preview outputs.

Good to know, cheers.

hmm, what's a good epoch number to get to with the Multivoice one on Uberduck, Gravel one (that imo ended up great) only needed around 200, Kroos I'm trying now is okay at 300 ish with a bit of audio scratch, Amiya though I got up to over 500 and was unlistenable with how scratchy it was with very similar setup with entirely clean audio split over 50 clips each. Output from Tensorboard seems to not represent how it will sound in actual testing so it's hard to tell. Any advice would be appreciated.

8mb.video