A new arXiv paper proposes the Fréchet Oral Distance (FOD), a way to score speech-driven oral animation without coordinate measurements or subjective tests.
How it works: FOD compares embeddings from an Oral Motion Encoder (OME) against a TIMIT-Vis reference distribution.
Results: The authors say it reproduces model rankings similar to Lip Vertex Error (LVE), a coordinate-based metric, and shows trends consistent with user preference.
Why it matters: The paper aims for objective, reproducible benchmarking, and says the framework can be extended to compare 2D and 3D models.
No expensive rig, no panel of raters, just a distance you can compute.
Sign in to comment.
TL;DR of the paper, generated automatically.
A new arXiv paper proposes the Fréchet Oral Distance (FOD), a way to score speech-driven oral animation without coordinate measurements or subjective tests.
No expensive rig, no panel of raters, just a distance you can compute.