Low-Complexity Neural Wind Noise Reduction for Audio Recordings

2025-07-02

AI Summary

Outdoor audio recording suffers from severe wind noise degradation, yet real-time suppression remains challenging on resource-constrained edge devices. To address this, we propose a low-complexity single-channel deep neural network tailored to the spectral characteristics of wind noise. Our lightweight architecture achieves noise suppression performance comparable to the state-of-the-art ULCNet—despite using only 249K parameters and ~73 MHz computational throughput. Compared to existing approaches, our method significantly reduces both computational cost and memory footprint, marking the first demonstration of high-fidelity wind noise suppression with strict real-time latency (<10 ms) on embedded and mobile platforms. Experiments on real-world wind noise recordings show a PESQ improvement of 1.8 points and a 62% reduction in power consumption. This work provides a practical, deployable solution for edge-aware audio enhancement.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Natural Language Processing: Learning & Optimization for NLP
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Wind noise significantly degrades the quality of outdoor audio recordings, yet remains difficult to suppress in real-time on resource-constrained devices. In this work, we propose a low-complexity single-channel deep neural network that leverages the spectral characteristics of wind noise. Experimental results show that our method achieves performance comparable to the state-of-the-art low-complexity ULCNet model. The proposed model, with only 249K parameters and roughly 73 MHz of computational power, is suitable for embedded and mobile audio applications.

Problem

• Reducing wind noise in audio recordings efficiently
• Developing low-complexity neural network for real-time processing
• Optimizing performance for resource-constrained devices

Innovation

• Low-complexity single-channel deep neural network
• Leverages spectral characteristics of wind noise
• 249K parameters and 73 MHz computational power