TOT-AWPE: A third-order tensor decomposition based algorithm for speech dereverberation

This repository is the official repository for “A third-order tensor decomposition based algorithm for speech dereverberation”.

📖 Paper

A third-order tensor decomposition based algorithm for speech dereverberation
Signal Processing, 239 (2026): 110210

🔗 https://doi.org/10.1016/j.sigpro.2025.110210

---

🔍 Overview

This work proposes TOT-AWPE, a third-order tensor decomposition based adaptive WPE algorithm, which:

• Decomposes the global multichannel prediction filter using a third-order Kronecker product structure
• Replaces the estimation of one long filter with multiple short, low-rank filters
• Achieves significant computational complexity reduction
• Maintains comparable or superior dereverberation performance compared to AWPE and KAWPE

📌 Citation

If you find this work useful, please cite:

@article{zhu2025third,
  title={A third-order tensor decomposition based algorithm for speech dereverberation},
  author={Zhu, Yujie and Huang, Gongping and Jin, Jilu and Chen, Jingdong and Benesty, Jacob},
  journal={Signal Processing},
  pages={110210},
  year={2025},
  publisher={Elsevier}
}