AI Sleep State Classification via Sound Signal Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional smart devices can only evaluate a user's sleep state based on activity levels and ambient light, failing to improve the sleeping environment by controlling peripheral devices effectively.
Innovation Solution
An AI apparatus that determines a user's sleep state through sound signal classification and controls home appliances to switch to a sleep mode, with adaptive learning for varying environments and user-specific optimization, and requests feedback for unclear detections to refine its model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional smart devices use activity level and ambient light to evaluate sleep state, then the evaluation can be performed, but the ability to control peripheral devices to improve sleep environment is insufficient
Solution Approach 1:
The patent combines multiple sensing modalities (sound signals from microphones, activity data from sensors, ambient light data) into a unified sleep state evaluation system. This integration allows the device to not only evaluate sleep state more accurately but also to control multiple peripheral devices (air conditioner, air purifier, lighting) based on the evaluated state, thereby resolving the contradiction between evaluation reliability and adaptability.
Solution Approach 2:
The AI apparatus is designed with multi-functional capabilities: it can evaluate sleep state, control peripheral home appliances, and adapt to individual user patterns. The system serves multiple purposes - monitoring, control, and personalization - which addresses the limitation of conventional devices that could only evaluate but not effectively control the sleep environment.
2Adaptability or versatility
If AI model uses sound signal classification to determine sleep state, then sleep environment control is improved, but the model needs continuous learning to handle varying user environments
Solution Approach 1:
The system implements feedback mechanisms where the AI model continuously learns from user responses and environmental data. When the model's sleep state determination is uncertain or when users provide feedback about their actual state, the system updates its learning parameters. This feedback loop enables the model to adapt to varying user environments without requiring complete retraining, balancing adaptability with manageable complexity.
Solution Approach 2:
The AI model performs preliminary learning during non-sleep periods to establish baseline user patterns and environmental characteristics. This preliminary action allows the model to be better prepared for accurate sleep state detection when the user actually sleeps, reducing the complexity of real-time decision-making while improving adaptability to individual users.
3Measurement precision
If the AI apparatus requests feedback for unclear detections, then learning accuracy is enhanced, but user interaction time increases
Solution Approach 1:
The system requests user feedback only in partial cases - specifically when the AI model's detection confidence is below a certain threshold or when environmental conditions are particularly challenging. For clear, confident detections, no feedback is requested. This selective feedback approach maintains high detection accuracy while minimizing the time users need to spend providing corrections.
Data Source
AI summary
A method for controlling an AI apparatus mounted on home appliance, the method comprising: receiving a sound signal at a predetermined time; removing noises of the received signal and separating the signal from which the noises are removed, into a signal by a user and a device signal; acquiring a result value outputted by an AI model by inputting the separated signal to the AI model using a multi-class separation; and executing a sleep mode of the home appliance and outputting a sleep mode switch notice when the result value is a user sleep, and requesting a feedback from the user when the result value is an unclear signal and updating the AI model using the feedback.