
Hi twittersphere, we’re Smule Labs, a research wing inside of @smule dedicated to exploring the frontiers of music, audio, and the human voice. We’ll be sharing some of our research here as it comes out 🧑💻

Charting the future of AI Audio

Hi twittersphere, we’re Smule Labs, a research wing inside of @smule dedicated to exploring the frontiers of music, audio, and the human voice. We’ll be sharing some of our research here as it comes out 🧑💻

Contributors who have X profiles: @Ray4Debasish @yongyi_zang

📝 Technical Blog & Live Experience Full engineering story: labs.smule.com/blog/inside-v… Try Inside Voice now (live for 2 weeks): labs.smule.com/x/vocal-tract…

🔧 Code: ctract We've open-sourced ctract, our differentiable inverse solver for @jamesondh's CHIMERA vocal tract synthesizer. MIT licensed: github.com/smulelabs/ctract

By replacing backpropagation through time with convolution, we achieved 47–63× faster vocal tract gradients on GPU.

We're also releasing our paper (ICASSP 2027 🚀), open-source code + technical blog. "Convolutional VocalTrax" D. R. Mohapatra, R. Leistikow, P. Cook arxiv.org/abs/2611.17181

Our first public Labs experiment is called Inside Voice. We wanted to visualize what’s actually happening inside your throat when you sing. It’s built on physics, not generative AI. Planning on keeping it live for a couple weeks. labs.smule.com/x/vocal-tract 🧵👇