AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Speaker-Adaptive Lip Reading with User-Dependent Padding

Minsu Kim, Hyunjun Kim, Yong Man Ro

ECCV 2022 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual speech separation · 28 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 user-dependent padding adaptation with 1 minute of dataLipNet-modified lip reading model · GRIDmean WER: 11.12 -> 6.80Improved
increase adaptation data from 1 to 5 minutesUser-dependent padding · GRIDmean WER: 6.80 -> 5.67Improved
add user-dependent padding adaptation with 1 minute of dataResNet-18 + MS-TCN · LRW-IDmean word accuracy: 85.85 -> 87.06Improved
increase adaptation data from 1 to 5 minutesUser-dependent padding · LRW-IDmean word accuracy: 87.06 -> 87.91Improved
add speaker-invariant adversarial training to user-dependent paddingUser-dependent padding · LRW-IDmean word accuracy (5min): 87.91 -> 88.48Improved
add speaker-invariant adversarial training to user-dependent paddingUser-dependent padding · GRIDmean WER (5min): 5.67 -> 5.80No clear effect
add speaker-invariant adversarial training to baselineBaseline lip reading model · GRIDmean WER: 11.12 -> 10.60Improved
add speaker code adaptation (1 minute) instead of paddingBaseline lip reading model · LRW-IDmean word accuracy: 85.85 -> 85.50Hurt
use 100% instead of 10% of adaptation dataUser-dependent padding · GRIDmean WER: 6.05 -> 4.65Improved
use 100% instead of 10% of adaptation dataUser-dependent padding · LRW-IDmean word accuracy: 87.35 -> 88.92Improved
finetune the whole model instead of padding, with less than 30% dataLip reading model adaptation · LRW-IDword accuracyHurt
finetune the whole model instead of padding, with more than 50% dataLip reading model adaptation · LRW-IDword accuracyImproved

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