TFMA-Net: A Time-Frequency Mamba Attention Network for Monaural Speech Enhancement
IEEE Signal Processing Letters · 2026 · 33 · 2660-2664
DOI: 10.1109/lsp.2026.3706070 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Hongfei Wang und 3 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Gefördert von: Natural Science Foundation of Shandong Province.
Abstract (Originalsprache)
Recent advances in speech enhancement have leveraged Transformer architectures for contextual modeling. Meanwhile, state-space models such as Mamba provide efficient long-sequence modeling but remain limited in capturing joint spectro-temporal correlations. To improve time–frequency (T–F) representation, we propose TFMA-Net, a time-frequency Mamba-attention network for speech enhancement. The proposed model integrates a Mamba module with an additional learnable linear residual path, multi-head self-attention (MHSA), and feed-forward networks within a unified block. Each block sequentially performs temporal and frequency modeling, enabling complementary dependency learning in both dimensions. Experimental results on the VoiceBank+DEMAND and DNS Challenge datasets show that TFMA-Net achieves competitive performance compared with recent attention- and Mamba-based baselines, while ablation results on TIMIT further verify the contribution of its submodules.
Förderung
- Natural Science Foundation of Shandong Province
Zitieren
APA
Wang, H., Gao, Y., Guo, X., & Ou, S. (2026). TFMA-Net: A Time-Frequency Mamba Attention Network for Monaural Speech Enhancement. IEEE Signal Processing Letters, 33, 2660-2664. https://doi.org/10.1109/lsp.2026.3706070
Vancouver
Wang H, Gao Y, Guo X, Ou S. TFMA-Net: A Time-Frequency Mamba Attention Network for Monaural Speech Enhancement. IEEE Signal Processing Letters. 2026;33:2660-2664. doi:10.1109/lsp.2026.3706070
BibTeX
@article{Wang2026,
author = {Wang, Hongfei and Gao, Ying and Guo, Xinyu and Ou, Shifeng},
title = {TFMA-Net: A Time-Frequency Mamba Attention Network for Monaural Speech Enhancement},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {2660--2664},
doi = {10.1109/lsp.2026.3706070}
}
EUVIMED