Acoustic Event Detection Using MFCCs and Mobile Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing acoustic event detection systems are complex, require significant equipment, and often need a priori information, making them inefficient for detecting specific events like gunshots, especially in noisy environments or at significant distances.
Innovation Solution
The use of Mel Frequency Cepstral Coefficients (MFCCs) to process sound information, compressing it into components on a log-frequency scale for comparison with an expected acoustic decay envelope, allowing for reliable detection of specific events like gunshots using mobile devices and ad hoc networks for geolocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple sensors located at significant distances from each other are used to detect the same acoustic event, then the detection coverage is improved, but the device complexity and equipment requirements increase significantly
Solution Approach 1:
The patent segments the acoustic detection task into multiple independent mobile devices, each performing local signal processing and event detection. Instead of requiring a centralized complex sensor array, each device independently analyzes acoustic events using MFCCs and template matching, distributing the detection function across multiple simple units that can operate autonomously or in coordination.
Solution Approach 2:
Each mobile device in the system performs self-service by independently generating MFCCs from received sound, comparing them against stored event templates, and making detection decisions locally. The devices do not require complex centralized processing or coordination, as each unit autonomously completes the full detection pipeline from signal acquisition to event identification.
2Loss of information
If Gaussian Mixture Models are used to detect a population of events, then the statistical analysis capability is improved, but the ability to detect specific individual events deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing template MFCCs for specific acoustic events (such as gunshots) before deployment. These templates represent the characteristic acoustic signatures of target events and are stored in each mobile device for rapid comparison. When an event occurs, the system compares the live MFCC against these pre-prepared templates, enabling fast and accurate specific event detection without requiring complex statistical modeling during the detection phase.
3Measurement precision
If a priori information is required for detecting specific acoustic events like gunshots, then the detection accuracy for known events is improved, but the adaptability to detect unknown or varying events deteriorates
Solution Approach 1:
The patent implements universality by designing a detection system that can handle multiple event types through a single unified approach. The same MFCC generation and template matching methodology works for detecting gunshots, fireworks, and other acoustic events. The system is not limited to a single event type but can detect various acoustic events by simply changing or adding the appropriate event templates, making it both accurate for specific events and adaptable to new event types.
Data Source
AI summary
Methods and apparatus to identify a type of acoustic event Mel Frequency Cepstral Coefficients (MFCCs). In one embodiment, received sound is processed using the MFCCs to locate a peak for comparison with a stored event to determine if the peak corresponds to the acoustic event.