60 epochs in ... can definitely make out some words now

what is this is how we learned to talk as babies?

idk dude maybe

ur brain just does stochastic gradient descent all day long

Update, she learned to speak english

BTW the reason the audio is so crap is because these samples are made with griffin-lim rather than a neural vocoder to save time while I'm training. I think I probably could make it better by denoising the griffin lim output but I haven't done that

Today is uberduck's highest day of usage ever by a lot, and I have no clue why

hello 🙂

what is your privacy policy about audio data sent to the duck?

hey! thanks so much for trying out the site

microphone data is not persisted after the session. The "download session" option that appears afterwards uses mic data which is from the client.

I'll put together a privacy policy for the site.

👌

Is uberduck.ai like vocodes?

We have a couple synthetic voices at https://uberduck.ai/speak, not as many as vocodes though. I'm trying to mainly focus on the interactive aspect at https://uberduck.ai but do plan to continue improving the voice synthesis

If you have any specific voice requests would be happy to work with you on that as well.

Yes

I have the data

Cool, what is the voice?

Michael stevens

I made a voice for 35 or 40 mins long

Awesome, will take a look

Thanks for putting it together

Yaay

Im gonna make a meme

You ping me if done

Vocodes doesnt accept our voices anymore

Hi

howdy!