Joint Detection and Active Cancellation of Snoring Signals in Real-Time

Snoring noise can be extremely annoying and troublesome, thus influencing people's social lives. Active noise control (ANC) systems can be implemented to reduce unwanted snoring noise, improving people's sleep quality, especially because ANC techniques are more suitable than passive solutions for low-frequency signals, like the snoring signal. However, the intermittent nature of snoring noise may affect the ANC performance. In this context, snoring activity detection (SAD) techniques can be applied to recognize the presence of snoring and accordingly activate the ANC system. The present work proposes a low-latency system based on deep learning for the detection of the snoring activity, combined with an active snoring cancellation (ASC) approach, which uses a delayless subband structure, for real-time applications. Experimental results have been obtained by evaluating the performance of the SAD algorithm, comparing two neural network-based solutions. Finally, the ASC system has been tested with and without the preliminary SAD stage, proving the advantage of using the SAD and ASC algorithms in a joint fashion.