Multi-channel Acoustic Event Detection via Power-Probability Image
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current acoustic event detection systems face challenges in detecting and classifying weak signals and false alarms due to the binary nature of voice activity detection, lack of multi-channel processing, and inability to consider neighboring channels, leading to missed events or excessive false alarms.
Innovation Solution
A two-level framework that generates a power-probability image by analyzing events in each channel independently and using a voting scheme, then classifies this image using machine learning to detect specific events or anomalies, incorporating power and probability tokens from multiple microphones to summarize contextual information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If binary voice activity detection is used to filter noise, then noise filtering is achieved, but weak acoustic events are eliminated or false alarms increase
Solution Approach 1:
The detection process is segmented into two stages: first stage performs independent single-channel event power and probability detection, second stage performs multi-channel power-probability image classification. This segmentation allows different detection strategies to be applied at different levels, reducing both false alarms and missed detections
Solution Approach 2:
The invention transitions from binary detection (single dimension) to power-probability joint detection (two dimensions). By creating a power-probability image that combines both dimensions across multiple channels, the system achieves more reliable detection without increasing false alarms
2Device complexity
If single-channel acoustic event detection is performed, then detection simplicity is maintained, but false alarms increase due to ignoring environmental context
Solution Approach 1:
The system segments the detection process into independent single-channel first stage and multi-channel collaborative second stage. This allows each channel to be processed independently initially, maintaining simplicity, while still enabling multi-channel context utilization for final classification
Solution Approach 2:
The power-probability image serves as an intermediary that consolidates information from multiple channels. This intermediary structure enables the system to consider environmental context from neighboring channels without directly processing complex multi-channel interactions, thus maintaining relative simplicity while reducing false alarms
3Reliability
If multi-channel processing with context analysis is implemented, then false alarms are reduced, but system complexity increases
Solution Approach 1:
The multi-channel processing is segmented into two distinct stages with different complexity levels. The first stage handles simple single-channel detection independently for each channel, while the second stage performs more complex power-probability image classification using accumulated events from multiple channels. This segmentation reduces overall system complexity compared to uniform multi-channel processing
4Measurement precision
If low detection thresholds are used to capture weak events, then event detection sensitivity increases, but false alarms increase
Solution Approach 1:
The system moves from single-threshold detection to two-dimensional power-probability space detection. By evaluating both power and probability dimensions simultaneously across multiple channels, the system can detect weak events with higher sensitivity while using the combined power-probability assessment to filter out false alarms that would occur with low single-threshold detection
Data Source
AI summary
A method for a multi-channel acoustic event detection and classification for weak signals, operates at two stages; a first stage detects a power and probability of events within a single channel, accumulated events in the single channel triggers a second stage, wherein the second stage is a power-probability image generation and classification using tokens of neighbouring channels.