@inproceedings{wang-etal-2026-bridging,
title = "Bridging the Temporal Gap in Multimodal {LLM}s: Deeply Stacking Temporal Tokens for Audio-Visual Speech Recognition",
author = "Wang, Liyong and
Xing, Junliang and
Hu, Tianyu and
Jiang, Jianfei and
Zhao, Jihuai and
Ma, Huimin",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.1381/",
doi = "10.18653/v1/2026.findings-acl.1381",
pages = "27748--27759",
ISBN = "979-8-89176-395-1",
abstract = "Audio-Visual Speech Recognition enhances speech recognition robustness in noisy conditions by leveraging visual cues. However, current Multimodal LLMs suffer from a fundamental temporal gap. This gap is characterized by limited fine-grained temporal modeling in vision encoders and progressive temporal semantic degradation throughout the deep layers of LLM decoders. To bridge this gap, we propose a novel framework that deeply stacks temporal tokens across both the encoding and decoding stages. Specifically, we enhance the vision encoder with a temporal-aware attention module and temporal rotary positional embeddings to precisely capture the sequential evolution and dynamics of lip movements. Furthermore, we stack hierarchical temporal tokens that incorporate temporally enriched features into multiple layers of the LLM decoder in a bottom-up manner. Extensive experiments on the LRS2 and LRS3 benchmarks demonstrate that our approach achieves high efficiency and firm performance, outperforming existing supervised, self-supervised, and LLM-based methods by 6.1{\%} on LRS2 and 7.8{\%} on LRS3."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-bridging">
<titleInfo>
<title>Bridging the Temporal Gap in Multimodal LLMs: Deeply Stacking Temporal Tokens for Audio-Visual Speech Recognition</title>
</titleInfo>
<name type="personal">
<namePart type="given">Liyong</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junliang</namePart>
<namePart type="family">Xing</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tianyu</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianfei</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jihuai</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huimin</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Audio-Visual Speech Recognition enhances speech recognition robustness in noisy conditions by leveraging visual cues. However, current Multimodal LLMs suffer from a fundamental temporal gap. This gap is characterized by limited fine-grained temporal modeling in vision encoders and progressive temporal semantic degradation throughout the deep layers of LLM decoders. To bridge this gap, we propose a novel framework that deeply stacks temporal tokens across both the encoding and decoding stages. Specifically, we enhance the vision encoder with a temporal-aware attention module and temporal rotary positional embeddings to precisely capture the sequential evolution and dynamics of lip movements. Furthermore, we stack hierarchical temporal tokens that incorporate temporally enriched features into multiple layers of the LLM decoder in a bottom-up manner. Extensive experiments on the LRS2 and LRS3 benchmarks demonstrate that our approach achieves high efficiency and firm performance, outperforming existing supervised, self-supervised, and LLM-based methods by 6.1% on LRS2 and 7.8% on LRS3.</abstract>
<identifier type="citekey">wang-etal-2026-bridging</identifier>
<identifier type="doi">10.18653/v1/2026.findings-acl.1381</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.1381/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>27748</start>
<end>27759</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Bridging the Temporal Gap in Multimodal LLMs: Deeply Stacking Temporal Tokens for Audio-Visual Speech Recognition
%A Wang, Liyong
%A Xing, Junliang
%A Hu, Tianyu
%A Jiang, Jianfei
%A Zhao, Jihuai
%A Ma, Huimin
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F wang-etal-2026-bridging
%X Audio-Visual Speech Recognition enhances speech recognition robustness in noisy conditions by leveraging visual cues. However, current Multimodal LLMs suffer from a fundamental temporal gap. This gap is characterized by limited fine-grained temporal modeling in vision encoders and progressive temporal semantic degradation throughout the deep layers of LLM decoders. To bridge this gap, we propose a novel framework that deeply stacks temporal tokens across both the encoding and decoding stages. Specifically, we enhance the vision encoder with a temporal-aware attention module and temporal rotary positional embeddings to precisely capture the sequential evolution and dynamics of lip movements. Furthermore, we stack hierarchical temporal tokens that incorporate temporally enriched features into multiple layers of the LLM decoder in a bottom-up manner. Extensive experiments on the LRS2 and LRS3 benchmarks demonstrate that our approach achieves high efficiency and firm performance, outperforming existing supervised, self-supervised, and LLM-based methods by 6.1% on LRS2 and 7.8% on LRS3.
%R 10.18653/v1/2026.findings-acl.1381
%U https://aclanthology.org/2026.findings-acl.1381/
%U https://doi.org/10.18653/v1/2026.findings-acl.1381
%P 27748-27759
Markdown (Informal)
[Bridging the Temporal Gap in Multimodal LLMs: Deeply Stacking Temporal Tokens for Audio-Visual Speech Recognition](https://aclanthology.org/2026.findings-acl.1381/) (Wang et al., Findings 2026)
ACL