Multi-Channel Speech Enhancement for Cocktail Party Speech Emotion Recognition

Youjun Chen ∗,1 , Guinan Li ∗,1 , Mengzhe Geng 2 , Xurong Xie †,3 , Shujie Hu 1 , Huimeng Wang 1 , Haoning Xu 1 , Chengxi Deng 1 , Jiajun Deng 1 , Zhaoqing Li 1 , Mingyu Cui 1 , Xunying Liu †,1 · 2026

Read Multi-Channel Speech Enhancement for Cocktail Party Speech Emotion Recognition by Youjun Chen ∗,1 , Guinan Li ∗,1 , Mengzhe Geng 2 , Xurong Xie †,3 , Shujie Hu 1 , Huimeng Wang 1 , Haoning Xu 1 , Chengxi Deng 1 , Jiajun Deng 1 , Zhaoqing Li 1 , Mingyu Cui 1 , Xunying Liu †,1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-18802.txt.

Description

At a cocktail party, recognising emotion means first pulling one voice out of a noisy mix. Multi-channel audio should help with that job, but it is rarely tested end to end. Youjun Chen, with eleven co-authors, shows in 2026 that a multi-channel speech enhancement front end materially improves speech emotion recognition in such settings.

A dereverberation and separation front end combining DNN-WPE with mask-based MVDR extracts the target speaker. A back end built on HuBERT and vision transformer features then judges emotion. On mixtures built from IEMOCAP and MSP-FACE data, the enhanced input beats fine-tuned single-channel baselines by clear margins in accuracy and F1 score. A front end trained on one dataset still generalises zero-shot to the other, which matters for real rooms.

Catalog record

CREATORYoujun Chen ∗,1 , Guinan Li ∗,1 , Mengzhe Geng 2 , Xurong Xie †,3 , Shujie Hu 1 , Huimeng Wang 1 , Haoning Xu 1 , Chengxi Deng 1 , Jiajun Deng 1 , Zhaoqing Li 1 , Mingyu Cui 1 , Xunying Liu †,1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-18802

Browse the full library · About the Tower of Babel