I'll try to help: Check if you put your wavs INSIDE the tacotron folder

so

it will be located in: /content/tacotron/wavs/

another thing is to check inside your transcript to see if the file directory is the same as the one inside the tacotron folder

i'll be making a tacotron training colab to help ease the pain of training

Here’s the problem, they are inside the tacotron folder. Everything’s where it supposed to be and yet its not recognizing anything. I’m worried that it could be the dataset itself but Im not sure what the problem would be, I made sure all the audio clips are mono at 22050 Hz.

@User I'm planning on making a google colab to help aid training for tacotron models, I'm planning on adding some cool features on it and keeping it up with weekly / daily maintenance. Would it be alright if you could recognize it as the official Uberduck colab?

Hmm, did you check if you had %cd /content/tacotron/ into the folder first before checking the files?

I did check and it used to be just /content, though once that got fixed it didnt fix much

I’ll show some screenshots that I sent to zwf

👍

for the no such file or directory you could right click on the list / transript and click copy path

I love this idea. we could link to it on the /about page.

👍 I'll start work

awesome

why does my tankman sound like a woman?

It's because when you train, you are mearly fine-tuning an already trained tacotron2 model. So it may sound similar to the original model than the actual speaker IIRC.

No but

it sounds like a woman exactly

did I train it for too long?

So does my voice with real time voice cloning for some reason

you may need more diverse datasets

/ more data in general

oki

Question. For a voice that has no more than two lines... it seems like the tacotron model already trained takes over for the voice.

There is minimum amount of dialogue for a voice to be individual from the trained one?

Yes, you need way more than two lines for training

Lemme rephrase that.

How many minutes do I need in a voice to at least be able to train a voice?