One of the ways that I believe people express their appreciation to the rest of humanity is to make something wonderful and put it out there.
San francisco
Joined February 2024
- Tweets25
- Following50
- Followers15
- Likes32
Have you ever needed a specific sound while making a video or a game?
It’s a model that lets you imitate a sound with your voice, then uses that vocal imitation together with text as input to generate the sound you actually want.
repo: github.com/thxxx/VTS
Have you ever needed a specific sound while making a video or a game?
It’s a model that lets you imitate a sound with your voice, then uses that vocal imitation together with text as input to generate the sound you actually want.
repo: github.com/thxxx/VTS
been looking at small NAR TTS models lately. Instead of DiT, they use convolution, so they’re much shallower (4–6 layers). Given speech’s nature, capturing local matters more than global, so conv seems to help a lot (would love to visualize attention maps for speech vs images)
Why is Vector Quantization needed in Auto-Regressive visual models?
I thinkg We can’t just train the model to regress the next token with L2 loss.
Because that assumes there's only one correct answer for every step—like saying "A cat is on the _" must always end with "floor".
But visual data (like language) is multimodal. There are many plausible next tokens. L2 loss pushes the model to output the average, killing diversity and expressiveness.
VQ turns this into a classification task over discrete codes, allowing the model to learn a distribution.
I like this guy's blog. It provides concise insights into deep learning.
nonint.com/
hojinkim retweeted
We’re OptimizerAI — creators of the highest-performing text-to-sound(effect) model.
We just launched our Discord community(discord.gg/vz53fDNwrs) — come join us!
We’ll be sharing exclusive discounts, upcoming updates, and more.
hojinkim retweeted
OptimizerAI v2 is HERE.
Every sound you’ve been searching for is right here—Generate the perfect sound to bring your vision to life with our LATEST model!
Now available at optimizerai.xyz
hojinkim retweeted
【効果音生成AIリリース！】
効果音がテキストから生成できる @OptimizerAI の1.0がリリース。動画に効果音があるだけでクオリティは1段階引き上げられる。
今後AIアニメプロジェクトでも試していきたいサービス。WebまたはDiscordで利用が可能
#生成AI #効果音
hojinkim retweeted
Introducing OptimizerAI 1.0 - the best text-to-sound effect model.
Bring your games, videos, movies, and animations to life with AI sound effects!
Available on web and Discord. Get started now at optimizerai.xyz/
hojinkim retweeted
Rolling out the web version and the high-performance v1 of OptimizerAI at the end of this month! 🚀 Eager to try text-to-sound right away? Come check it out on optimizerai.xyz 🎶✨
hojinkim retweeted
What if there are sound effects in AI videos..? @pika_labs Lip Sync met @OptimizerAI Sound Effects
Hey everyone!
I'm excited to introduce what I'm making today. It's a free AI tool that allows you to generate sound effects for your games just by using text! 🔥💪🏻
I hope you all find this tool as helpful as we do 😃
Please show us lots of support! 🥳
#gamedev #indiegamedev