Skip to content
Open access

Efficient speech denoising using CleanUNet optimized with Mamba and hybrid spectral loss

Aug 2026 · Multimedia tools and applications · Vol 85 · 0 citations · 23 references

TL;DR

An optimized version of CleanUNet is proposed that significantly reduces model complexity while maintaining competitive performance, and replaces the original Transformer bottleneck with a Mamba module, which offers efficient sequence modeling with linear time complexity.

Abstract

Speech enhancement aims to recover clean speech from signals degraded by noise and adverse acoustic conditions, such as background interference and echo. CleanUNet has emerged as an effective solution for causal speech denoising, but its high computational demands limit deployment on resource-constrained devices. In this work, we propose an optimized version of CleanUNet that significantly reduces model complexity while maintaining competitive performance. We replace the original Transformer bottleneck with a Mamba module, which offers efficient sequence modeling with linear time complexity. Additional architectural improvements include batch normalization, lightweight channel attention, alternative activation functions, and a novel hybrid loss function that combines Short-Time Fourier Transform and Wavelet-based components. Our best models achieved up to 86% parameter reduction relative to the original CleanUNet baseline with minimal performance degradation, and an ultralight version reduced parameters by up to 98% while still delivering competitive results in terms of Perceptual Evaluation of Speech Quality (PESQ) and Short-Time Objective Intelligibility (STOI). A comprehensive Pareto analysis illustrates the trade-offs between model size and speech quality, providing a range of configurations suited for different application scenarios, including real-time and embedded environments.

Read PDF

Similar papers

Preprint Aug 2026

A Hybrid Classical-Learning Framework for Adaptive Decision Directed Speech Enhancement

An Adaptive Beta-Constrained Decision-Directed (ABCDD) speech enhancement framework that extends the conventional DD method through a frame-dependent lower gain bound and combines interpretable classical enhancement structure with lightweight machine-learning-based parameter adaptation provides an effective and practic...

Ali Rajabi, Xiang-Wei Zhou · 0 citations
Open access Aug 2026

A deep residual complex learning framework with long-range temporal context for phase-aware speech enhancement

Analysis of the results indicates that this multi-objective optimization framework achieves superior speech enhancement by simultaneously addressing magnitude and phase reconstruction, waveform-level distortion, and psychoacoustic naturalness, a balance not achieved in any prior works so far.

Debabrata Gogoi, Sushanta Kabir Dutta · 0 citations
Sep 2026

A dual-stream spectral network for versatile full-band speech enhancement.

Evaluation results demonstrate that DSVN improves restoration quality under both single- and mixed-degradation scenarios, and shows clear advantages in non-intrusive quality assessment, spectral distance, and automatic speech recognition evaluation, indicating that the enhanced speech is not only cleaner but also more...

Jun-Kang Yang, Hiromitsu Nishizaki, C. Leow et al. · 0 citations
Review Open access Sep 2026

Advances in Speech Enhancement: A Comprehensive Review of Noise Suppression Techniques

Overall, this review provides a unified perspective on speech enhancement techniques, identifies their strengths, and highlights emerging research opportunities for the development of robust, efficient, and intelligent speech enhancement systems.

Pushpraj Tanwar, A. Somkuwar, Rakesh Kumar Gumasta · 0 citations
Open access Sep 2026

Machine Learning-Enhanced Uniform DFT Polyphase Filter Bank for Real-Time Speech Enhancement in Health Care Applications

Purpose – To improve speech quality and intelligibility in noisy healthcare environments, such as ICUs, emergency wards, and telemedicine systems, using a machine learning-enhanced speech enhancement framework. Design/methodology/approach – The proposed method uses a Uniform DFT Polyphase Filter Bank (PFB) to divide no...

Thaseena C. K., G. Sodhi · 0 citations
Open access Aug 2026

Noise Reduction Strategies For Optimising Speech Signal Clarity

Experimental results indicate that CSM accelerates the training of the models while enhancing the speech quality and show that the method effectively denoises noisy speech audio while being both efficient and accurate.

S. More, J. Helonde, P. Burade · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.