Recent submissions

Comment · 2 years ago
Comment · 2 years ago
Comment · 3 years ago
love the deep dive here
Comment · 3 years ago
Those are awesome! Sorry if the app is down for you or anyone else right now btw. Our servers are a little overloaded but we'll be back up in no time
Comment · 3 years ago
We would have an iOS app if we were better mobile engineers... hopefully someday we'll make one!
Comment · 3 years ago
Very cool. The latent space is a wild place.
Comment · 3 years ago
Nice! It is best at pronouncing in English, but we've had a bunch of fun trying to get other languages too. Sometimes you can make things happen phonetically.

Even for english words that it doesn't get right the first time haha

Comment · 3 years ago
We're happy to be building a toy!

Your comment reminds me of this post: https://cdixon.org/2010/01/03/the-next-big-thing-will-start-...

It's still really early innings for this technology, so we're happy to be learning and building fun technology helps people to do creative things. Main thing we're focused on is turning it from a toy you enjoy once, to something you come back to and dive deeper into (even if it's still just for fun).

Comment · 3 years ago
Comment · 3 years ago
Would be neat to see the distribution of time that people stare at this... I imagine some will stare for hours
Comment · 3 years ago
Author here: fwiw we are running the app on a10g GPUs, which generally can turn around a 512x512 in 3.5s with 50 inference steps. This time includes converting the image into audio which should be done on the GPU as well for real-time purposes. We did some optimization such as a traced unet, fp16 and removing autocast. There are lots of ways it could be sped up further I'm sure!
Comment · 3 years ago
Author here: We were blown away too. This project started with a question in our minds about whether it was even possible for the stable diffusion model architecture to output something with the level of fidelity needed for the resulting audio to sound reasonable.
Comment · 3 years ago
Author here: It can certainly be applied to voice, but the model would need deeper training to speak intelligibly. If you want to hear more singing, you can try a prompt like "female voice", and increase the denoising parameter in the settings of the app.

That said, our GPUs are still getting slammed today so you might face a delay in getting responses. Working on it!

Comment · 3 years ago
Author here: Indeed we are using Griffin-Lim. Would be exciting to swap it out with something faster and better though. In the real-time app we are running the conversion from spectrogram to audio on the GPU as well because it is a nontrivial part of the time it takes to generate a new audio clip. Any speed up there is helpful.
Comment · 3 years ago
Authors here: Fun to wake up to this surprise! We are rushing to add GPUs so you can all experience the app in real-time. Will update asap

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton