AI Paper Atlas

Research map / 3D vision / Face and shape

MemoryTalker: Personalized Speech-Driven 3D Facial Animation via Audio-Guided Stylization

Hyung Kyu Kim, Sangmin Lee, Hak Gu Kim

ICCV 2025 · 3D vision › Face and shape · approach: Speech-driven 3D facial animation · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add 1-stage motion memory training (ablation)MemoryTalker baseline · VOCASETLVE (x1e-5): 0.460 -> 0.313Improved
add 1-stage motion memory training (ablation)MemoryTalker baseline · VOCASETFVE (x1e-6): 0.638 -> 0.531Improved
add 2-stage stylized motion memory (ablation)MemoryTalker 1-stage · VOCASETLVE (x1e-5): 0.313 -> 0.293Improved
add 2-stage stylized motion memory (ablation)MemoryTalker 1-stage · VOCASETFVE (x1e-6): 0.531 -> 0.506Improved
add lip vertex loss L_lip (ablation)MemoryTalker · VOCASETLDTW (x1e-5): 0.435 -> 0.418Improved
add triplet speaking style loss L_style (ablation)MemoryTalker · VOCASETLDTW (x1e-5): 0.431 -> 0.418Improved
use two-stage instead of end-to-end training (ablation)MemoryTalker · VOCASETFID (x1e-1): 3.142 -> 3.045Improved
use mel-spectrogram instead of ASR features for style encoder (ablation)MemoryTalker · VOCASETLVE (x1e-5): 0.300 -> 0.293Improved
use audio-only style instead of mesh-driven styleMemoryTalker · VOCASETFVE (x1e-6): 0.673 -> 0.506Improved
vary memory slot size 16, 24, 32, 48, 64 (ablation)MemoryTalker · VOCASETLVEMixed
use MemoryTalker instead of UniTalkerUniTalker · BIWI Test-BLVE (x1e-4): 0.196 -> 0.187Improved
use MemoryTalker on seen identities instead of UniTalkerUniTalker · BIWI Test-A (seen identities)LVE (x1e-3): 0.283 -> 0.266Improved

Later papers that compared with this method

1 later accepted paper reports MemoryTalker as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy