Reproduce “Reconstructing the Vocal Tract with Differentiable Acoustic Simulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Eric Ming Chen, Jin Woo Lee, Vincent Sitzmann
The authors reconstruct vocal-tract shapes and MRI sequences from speech by making a physical tube simulator differentiable, then optimizing candidate area functions—cross-sectional areas along the tract—to match the audio. A neural-field representation regularizes this ill-posed inverse problem. On 40 three-second sequences, their MRI method reports FVD 623 versus 2,949 for Speech2rtMRI, but anatomy is not uniquely determined.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.