Real-Time Wind Noise Detection and Suppression with Neural-Based Signal Reconstruction for Mult-Channel, Low-Power Devices
This work addresses wind noise reduction for ASR functionality in smart glasses and IoT devices, presenting an incremental improvement with specific algorithms for low-power settings.
The paper tackled wind noise interference in speech recognition for wearable devices by developing a real-time detection algorithm and a neural-based suppression method, achieving high accuracy in detection and effective speech signal reconstruction under various wind conditions.
Active wind noise detection and suppression techniques are a new and essential paradigm for enhancing ASR-based functionality with smart glasses, in addition to other wearable and smart devices in the broader IoT (Internet of things). In this paper, we develop two separate algorithms for wind noise detection and suppression, respectively, operational in a challenging, low-energy regime. Together, these algorithms comprise a robust wind noise suppression system. In the first case, we advance a real-time wind detection algorithm (RTWD) that uses two distinct sets of low-dimensional signal features to discriminate the presence of wind noise with high accuracy. For wind noise suppression, we employ an additional algorithm - attentive neural wind suppression (ANWS) - that utilizes a neural network to reconstruct the wearer speech signal from wind-corrupted audio in the spectral regions that are most adversely affected by wind noise. Finally, we test our algorithms through real-time experiments using low-power, multi-microphone devices with a wind simulator under challenging detection criteria and a variety of wind intensities.