Use of IoT-based Solution and Audio Data Analysis for Snoring Detection

Snoring is a common sleep phenomenon whereby loose tissue in the human air passages vibrates as the sleeper inhales. In most circumstances, snoring is quite normal; but snoring which is particularly loud and frequent can be indicative of underlying health complications, such as Obstructive Sleep Apnea (OSA). Snorers that share sleeping quarters with other people can be a nuisance and negatively affect the sleep of others, but they can be informed that they snore and that they should seek medical attention. On the other hand, individuals that live and sleep on their own might not be aware that they snore while they sleep, complicating any medical intervention that could treat the problem. This paper presents the development of a hardware-software hybrid solution to detect snoring using Deep Learning (DL) techniques, including the discussion of system architecture, design, and initial results. A DL model is trained using a dataset of sound files of snoring and non-snoring noise. The proposed IoT-based solution is tested on sound files of snoring and non-snoring from an online dataset, and on sound files recorded of a human tester through the Single-Board Computer (SBC) used. For the snoring detection, the sound files are processed into a spectrogram which the model can parse and sent to the DL model to be classified into snoring or not-snoring, which is then presented to the user in the form of a basic dashboard interface created in Python. The results obtained are promising with the highest values being 99.4% accuracy, 99% recall, 99.7% precision, and 99% specificity.