SnoringNoPlease

The Inspiration & The Problem

When exploring the mysteries of sleep, especially during the deep hours of the night when our bodies are entirely at rest changes in our health and breathing patterns are commonly heard much more easily than they are seen. For the average person, monitoring sleep quality using visual cues or cameras in a pitch-black bedroom is impractical and raises massive privacy concerns. Yet, ignoring what happens in the dark can be dangerous.

Sleep apnea is a common sleep disorder that causes repeated interruptions in breathing during the night, often remaining undiagnosed because symptoms occur while the person is asleep. Left untreated, it can lead to serious health problems such as chronic fatigue, cardiovascular disease, reduced concentration, and a significantly lower quality of life. While clinical diagnostic methods like polysomnography (sleep studies) are highly accurate, they are also expensive, uncomfortable, and not easily accessible to everyone.

Our Solution: Audio-Based Edge AI

The most prominent and telling feature of our nighttime environment is actually sound. This makes audio-based classification a much more practical, completely non-invasive approach for preliminary screening.

To help improve early detection and make monitoring accessible, we developed SnoringNoPlease. Powered by an acoustic Machine Learning model trained and deployed via Edge Impulse, this bedside system is capable of identifying breathing irregularities and potential apnea episodes during sleep in real time.

Machine Learning & Model Training

We used the Edge Impulse platform to train our machine learning model on audio samples, focusing on effectively identifying snoring patterns and distinguishing them from environmental background noise.

To build a robust dataset, we collected and filtered clear, distinctive audio clips of snoring. We also intentionally included ambient background noise (such as fans, air conditioning, or general room acoustics) without any snoring. This ensures that our model can accurately detect irregularities in real-world home environments without being fooled by everyday household sounds.

As shown in the training performance metrics above, our model achieved an impressive 90.4% overall accuracy on the validation set. It demonstrates an excellent 96.4% true positive rate for detecting snoring and a strong F1-score of 0.92, proving it is highly reliable for a non-invasive assistant.

The Data Explorer visualization highlights how cleanly the model separates actual snoring features from baseline room noise, ensuring high confidence during real-time overnight monitoring.

Web Application Architecture

Since many sleep monitoring systems often limit the user to specific, expensive hardware or uncomfortable wearable sensors, we wanted to make our app as widely accessible, lightweight, and universal as possible. It features a modern, responsive dashboard layout built with React, Tailwind CSS, and Shadcn UI components that seamlessly adapts to any device—whether the user places a smartphone or a laptop on their bedside table.

How the Application Works

https://snore-sense-lab.netlify.app/

The user workflow is incredibly simple, streamlined, and designed for overnight usage:

One-Tap Activation: Before going to sleep, the user simply clicks the microphone button on the dashboard to start the session

Continuous Overnight Monitoring: The application actively listens through the device microphone all night long, analyzing incoming audio buffers in real-time.

• Smart Snoring Classification: As audio streams in, the local machine learning model instantly categorizes periods of SNORING versus baseline ambient NOISE
• Intelligent Interval Tracking: Crucially, the app doesn't just count snorings; it actively measures the time elapsed between each snoring event. Tracking these quiet gaps and interruptions is critical for highlighting irregular breathing patterns that might point to potential sleep apnea episodes.

Future improvements

In the future, we plan to continue improving our web application by adding more accurate snoring detection, better visual reports, and more personalized recommendations for users. We also aim to collaborate with companies that develop sleep-improvement products, as well as official medical organizations, in order to make the solution more reliable and useful from a health perspective.

Another future goal is to integrate our model into smart wearable devices, such as smart bracelets or watches. These devices could monitor snoring during sleep and gently wake the user when abnormal or potentially harmful snoring patterns are detected. This would help users react in real time and could contribute to better sleep quality and improved health monitoring.