AI Paper Atlas

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

Mutual Learning for Acoustic Matching and Dereverberation via Visual Scene-driven Diffusion

Jian Ma, Wenguan Wang, Yi Yang, Feng Zheng

ECCV 2024 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual source localization · 3 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 mutual learning between reverberator and dereverberator (no unpaired data) (ablation)VSD · SoundSpaces-Speech (VAM)STFT distance: 0.657 -> 0.573Improved
add unpaired natural and anechoic audio to mutual learning (ablation)MVSD · SoundSpaces-Speech (VAM)STFT distance: 0.573 -> 0.508Improved
add unpaired data to mutual learning for dereverberation (ablation)MVSD · SoundSpaces-Speech (dereverberation)WER (%): 4.32 -> 4.27Improved
add visual scene conditioning (ablation)MVSD · SoundSpaces-Speech Seen (VAM)RTE (s): 0.188 -> 0.030Improved
replace GAN-based generator with diffusion model (ablation)MVSD · SoundSpaces-Speech (VAM)STFT distance: 0.831 -> 0.657Improved
add diffusion process to controllable Unet (ablation)MVSD · SoundSpaces-Speech (dereverberation)WER (%): 6.74 -> 4.27Improved
increase denoising steps from 150 to 250 (ablation)MVSD · SoundSpaces-Speech (VAM)STFT distance: 1.452 -> 0.508Improved
increase denoising steps from 250 to 1000 (ablation)MVSD · SoundSpaces-Speech (VAM)STFT distance: 0.508 -> 0.487No clear effect
increase unpaired data from 0k to 5k (ablation)MVSD · SoundSpaces-Speech (VAM)STFT distance: 0.573 -> 0.508Improved
add mutual learning (MVSD) vs SOTA AViTAR on Seen splitVAM · SoundSpaces-Speech Seen (VAM)STFT distance: 0.665 -> 0.508Improved

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