Super Smash Bros Ultimate Announcer | 20 Epochs | GPT-SoVITS

[river flower]

https://huggingface.co/Plasmati/AIDRIVEORNOTMODELS8/resolve/main/SSBU-Announcer.zip?download=true

[feral relic]

Rate the model

To rate this model, either use /rate or click the buttons below.

[muted gulch]

Where’s the Transcript and Reference audio?

[river flower]

The reference audio is in the zip

Damn 1 of rating is it that bad

[muted gulch]

Ok anyways W Model

[river flower]

Also bcuz I don't know about copyright when I uploaded the data for a family guy model Pikachu told me to not do that so I'm avoiding discord uploads when it comes to data or reference

[river flower]

Long ahh message

[river flower]

Nah

Rvc

Ye I guess

[scarlet holly]

uh, better than putting the audio reference there

btw, I wish that audio references inside the zip were mandatory.

[feral relic]

https://cdn.discordapp.com/attachments/1160576561636966491/1165372018414342284/warning.gif

[empty shore]

but why? isn't more easier to publish the ref + transcripts directly?

[scarlet holly]

1- There are many GPT-SoVITS models without audio reference. (Most of them don't have reviews because there's no audio reference for them.)
2- You will need to ask them for an audio reference. (If they dont post one)
3- The audio reference will be already included in the zip file without having to download the zip file and the audio reference at the same time.
4- One audio reference wouldn't be enough to make inferences (most people only share 1 audio reference), because if you want to make a model do a certain emotion, you need to get other audio references with different emotions.
5- Posting an audio reference in the forum would only make things more difficult and would cause the chat to be flooded if the person is sharing 2 or more audio references.

[empty shore]

so that's why ur gpt sovits refs and transcriptions are always on a zip...

[scarlet holly]

yup

[empty shore]

i will keep in mind putting ref audios + transcriptions on zip if there's a case where i have to do 2 or more audio references

[scarlet holly]

Also, a note to number 2: This would make your model less famous, because it would be impossible to use unless you search for datasets on the Internet (I think no one would do that).

[empty shore]

Or people can be lazy doing audio refs + transcriptions manually

[scarlet holly]

I usually put all the sliced audios as audio references. I don't care if someone merges all the sliced audios and steals the dataset.

[empty shore]

I do ref audios manually

Recycling the dataset used for training

[scarlet holly]

hold the truth

https://tenor.com/view/trisha-trisha-paytas-h3-h3h3-h3podcast-gif-23747167

[river flower]

Real

[near echo]

@river flower where's the reference audio?

[scarlet holly]

in the zip