# Reverse Attention for Lightweight Speech Enhancement on Edge Devices

arXiv: https://arxiv.org/abs/2509.16705 · HTML: https://mlsanity.com/paper/2509.16705/

## Key facts

- No citing papers in the map yet (data through 19 Oct 2026).
- Placed in Speech enhancement and noise suppression, one of 11 directions in Speech and audio language models.
- Builds on 8 of its 15 references that are in the map.
- First posted on arXiv 20 Sep 2025.

## Placed in

- Speech and audio language models / [Speech enhancement and noise suppression](https://mlsanity.com/speech/speech-enhancement-and-noise-suppression/)

## Builds on (most cited first, 8 of 8)

- [Attention U-Net: Learning Where to Look for the Pancreas](https://mlsanity.com/paper/1804.03999/) (2018)
- [StoRM: A Diffusion-based Stochastic Regeneration Model for Speech Enhancement and Dereverberation](https://mlsanity.com/paper/2212.11851/) (2022)
- [DeepFilterNet: A Low Complexity Speech Enhancement Framework for Full-Band Audio based on Deep Filtering](https://mlsanity.com/paper/2110.05588/) (2021)
- [Schr"odinger Bridge for Generative Speech Enhancement](https://mlsanity.com/paper/2407.16074/) (2024)
- [GenSE: Generative Speech Enhancement via Language Models using Hierarchical Modeling](https://mlsanity.com/paper/2502.02942/) (2025)
- [Semantic Segmentation with Reverse Attention](https://mlsanity.com/paper/1707.06426/) (2017)
- [Towards Sub-millisecond Latency Real-Time Speech Enhancement Models on Hearables](https://mlsanity.com/paper/2409.18239/) (2024)
- [Speech enhancement deep-learning architecture for efficient edge processing](https://mlsanity.com/paper/2405.16834/) (2024)

## Abstract (arXiv)

This paper introduces a lightweight deep learning model for real-time speech enhancement, designed to operate efficiently on resource-constrained devices. The proposed model leverages a compact architecture that facilitates rapid inference without compromising performance. Key contributions include infusing soft attention-based attention gates in the U-Net architecture which is known to perform well for segmentation tasks and is optimized for GPUs. Experimental evaluations demonstrate that the model achieves competitive speech quality and intelligibility metrics, such as PESQ and Word Error Rates (WER), improving the performance of similarly sized baseline models. We are able to achieve a 6.24% WER improvement and a 0.64 PESQ score improvement over un-enhanced waveforms.