AI Paper Atlas

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

Neural-Inspired Modeling of Auditory Selection and Compensation for Audio-Visual Speech Separation

Xinmeng Xu, Haoran Xie, Xiaohui Tao, Lin Li, S. Joe Qin

ICML 2026 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual speech separation

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 auditory selection module (ASM) (ablation)Neuro-SCNet · LRS2-4MixSI-SNRi (dB): 8.6 -> 9.8Improved
add auditory selection module (ASM) (ablation)Neuro-SCNet · LRS2-2MixSI-SNRi (dB): 16.4 -> 17.2Improved
add cross-modal compensation module (CCM) (ablation)Neuro-SCNet · LRS2-4MixSI-SNRi (dB): 8.4 -> 9.8Improved
add cross-modal compensation module (CCM) (ablation)Neuro-SCNet · LRS2-2MixSI-SNRi (dB): 16.2 -> 17.2Improved
add both ASM and CCM (ablation)Neuro-SCNet · LRS2-4MixSI-SNRi (dB): 7.0 -> 9.8Improved
use estimated phase instead of mixture phase (ablation)Neuro-SCNet · LRS2-4MixSI-SNRi (dB): 6.5 -> 9.8Improved
use oracle clean phase instead of learned phase (ablation)Neuro-SCNet · LRS2-2MixSI-SNRi (dB): 17.2 -> 17.5Improved
add audio encoder alongside joint encoder (ablation)Neuro-SCNet · LRS2-4MixSI-SNRi (dB): 8.1 -> 9.8Improved
use parallel magnitude and phase decoders instead of single-domain modeling (ablation)Neuro-SCNet · LRS2-3MixSI-SNRi (dB): 11.5 -> 13.9Improved
add visual pre-alignment offset estimation and reliability weighting (ablation)visual path · LRS2 at 200 ms shiftSI-SNRi (dB): 14.8 -> 16.2Improved
add reliability weighting to offset-only alignment at zero shift (ablation)visual path · LRS2 at 0 ms shiftSI-SNRi (dB): 16.9 -> 17.2Improved
use Neuro-SCNet instead of IIANet with four speakersIIANet · LRS2-4MixSI-SNRi (dB): 8.2 -> 9.8Improved

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