Smart Anti Snoring AI Device
Platform
ESP32-S3
Type
Embedded AI / Health
Model
CNN + INT8
Status
Completed
Background
Privacy-First On-Device AI for Real-Time Snore Detection Without Cloud Dependency
The On-Device Snore Detection System is an innovative solution designed to address the growing need for affordable, non-invasive, and privacy-preserving methods for monitoring sleep-related disorders. Traditional diagnostic approaches, such as clinical sleep studies, are expensive, intrusive, and impractical for long-term use in a home environment. The project was developed to provide a real-time, embedded system that could detect snoring events locally on a resource-constrained device, ensuring privacy by eliminating the need for cloud connectivity.
Challenges
Key Project Challenges
1
Limited Hardware Resources
Running a real-time AI system on a microcontroller with constrained flash memory, RAM, and processing power without sacrificing performance.
2
Accuracy vs. Efficiency
Achieving high classification accuracy of approximately 91% while keeping the model lightweight enough to run offline in real time.
3
Privacy Concerns
Processing all audio data entirely on-device to eliminate cloud dependency and protect sensitive user sleep data at all times.
4
Continuous Embedded Operation
Designing the system to run indefinitely without excessive battery drain or hardware resource consumption during overnight monitoring.
Solutions
How We Built It
Our Approach
Optimized On-Device CNN with INT8 Quantization on ESP32-S3
A custom Convolutional Neural Network (CNN) was developed and optimized for embedded deployment on the ESP32-S3 microcontroller. All audio processing, including feature extraction and inference, runs entirely on-device with no cloud dependency. The model was optimized using global average pooling, progressive dropout, and controlled channel growth to prevent parameter explosion, while INT8 quantization reduced the model size without significantly impacting accuracy. The result is a system achieving approximately 91% classification accuracy with low inference times, minimal power consumption, and continuous real-time performance in a home environment.
Benefits
Value Delivered
Improved Accessibility
Provides a low-cost, home-based alternative to clinical sleep studies, making sleep disorder monitoring widely accessible.
Privacy Preservation
All audio data is processed entirely on-device, ensuring sensitive sleep data never leaves the user's home environment.
Real-Time Performance
Delivers reliable, continuous snore detection at approximately 91% accuracy, suitable for uninterrupted overnight home use.
Scalability & Market Potential
Technology is scalable to consumer health and wearable products, with potential to reduce reliance on clinical initial screenings.
Validated Market Interest
An Indiegogo campaign generated early-stage market interest and real-world user feedback to guide further product development.
Client Feedback
What the Client Said
"
The privacy angle was what made this project difficult — our users were not going to accept their sleep audio being sent anywhere. Getting 91% accuracy running entirely offline on a microcontroller is something we genuinely didn't think was achievable at this price point. The model is compact, the battery holds up through the night, and the Indiegogo response confirmed we had something real. This is exactly the kind of embedded AI work that's hard to find.
Have a Similar Project?
We build custom embedded AI and hardware solutions tailored to your needs.
Request a Free Quote →