im gonna try training a model based on 6 seconds of data (it's all that exists lol) wish me luck hfdsjk

soo how was it?

It turned out okay I trained it on the arpabet notebook and the pipeline notebook The arpabet one worked fine. It can't synthesize well beyond 2 seconds, and because of lack of data it can't say some words The pipeline one only says like "are you ready" over and over again. So like... "are you ready arrrrre you rrreadyready irone amm irone aka" on longer sentences

I meant to send samples here but i forgot

I want to try to train it on two more notebooks still And i also want to try to make a talknet out of it But if i still get results like that i think i'll start tuning more samples for the dataset. Basically, the dataset is the english part of the exvoice samples for an utau called irone aka. I can make more samples from the voicebank itself, but those samples would be less natural ofc since they'd be made by me rather than pre-recorded

gaugan 2

1. input = "city landscape"

bruh

2. input = "city factory" instead it shows me an image without a factory

Wrong channel, #841461499946860576

showcasing a machine learning website bruh

Oh, sorry😳

Irone Aka 7-wav test (6 seconds of data) "How are you?."

"I am here."

"This essay is divided into three segments[i dont remember if that's the right word, i didnt write it down]"

Kotonoha Akane 15-wav test "The totality that encompasses the two is the world as idea, or the phenomenon."

"Only knowledge remains, the will has vanished."

that's very impressive

It was an overtrain

It was a perfect save before, but then my PC lagged and it got enough time to save at a worsened spot later

0.85 val loss when I wanted to save

1.1 val loss when it saved later

WAIT no

0.085* and 0.11*

you might still be able to download an older version and reupload it to drive

How low can you get the number for the model down to?

Like I have it at 0.10 right now

Also, what is that number called?

I got it to 0.095 so I stopped it there and I'm gonna see how it sounds

@Liam Taheny I had been working on that Ben Patrick Johnson voice \:) So far, sometimes it turns out really good but other times it does not. Mostly because I used different clips which varied from a more spacious reverby room and others that were him actually doing a voice over. Check out this one though which was generated by the model: