Sound Event Detection with Event-Level Thresholding

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional sound event detection systems rely on frame-level thresholding, which entangles event boundary and confidence information, leading to false alarms and missed hits due to subjective threshold choices affecting event predictions.

Innovation Solution

Implementing event-level thresholding through sound event bounding boxes (SEBBs) to decouple temporal extent and confidence prediction, ensuring that changing thresholds do not alter event spans, and using post-processing techniques to convert frame-level scores into SEBBs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If frame-level thresholding is used to control system sensitivity, then the ability to detect events is improved, but the precision of event boundary detection deteriorates because the threshold affects both event presence and event extent

Engineering Contradiction:

Improveevent detection reliabilityVSAvoidevent boundary precision

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The patent segments the event detection process into two independent components: event presence detection (using frame-level thresholding) and event extent determination (using dynamic programming to find optimal boundaries). This segmentation allows each component to be optimized independently, resolving the contradiction between sensitivity control and boundary precision.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary optimization step (dynamic programming) that mediates between the frame-level presence decisions and the final event boundaries. This intermediary process refines the raw frame-level detections into precise event extents without being affected by the threshold choice, thus decoupling the two previously entangled functions.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Productivity

If a lower frame-level threshold is used to increase sensitivity and detect more events, then the recall is improved, but false alarms increase and event boundaries become less accurate

Engineering Contradiction:

Improveevent detection recallVSAvoiddetection accuracy

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

By segmenting the detection process into presence detection and boundary optimization, the system can use low thresholds to maximize recall in the presence detection phase, while the subsequent boundary optimization phase eliminates false alarms by enforcing temporal consistency and selecting only the most reliable event boundaries.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The dynamic programming optimization provides feedback to the frame-level detections by evaluating which detected events satisfy temporal consistency constraints and boundary optimization criteria. This feedback mechanism filters out false positives that would otherwise be introduced by low thresholding, maintaining high reliability while preserving high recall.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If a higher frame-level threshold is used to reduce false alarms, then the precision is improved, but missed hits increase

Engineering Contradiction:

Improvedetection precisionVSAvoidevent detection recall

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system performs preliminary detection at high sensitivity (low threshold) to ensure all potential events are captured, then applies preliminary filtering through the boundary optimization process that removes false positives. This preliminary action-preliminary filtering sequence achieves both high recall and high precision, resolving the contradiction between using high or low thresholds.

Inventive Principle:

Principle #10Preliminary action

4Device complexity

If frame-level thresholding is applied to each frame independently, then the detection process is simplified, but the event extent prediction becomes entangled with confidence information

Engineering Contradiction:

Improvedetection process complexityVSAvoidseparation of extent and confidence information

Core Design Contradiction:

Device complexityVSLoss of information

Solution Approach 1:

The patent segments the detection output into separate confidence scores (from frame-level thresholding) and extent information (from dynamic programming optimization). This segmentation preserves both pieces of information independently, allowing confidence to reflect detection certainty while extent reflects temporal boundaries, without the entanglement that occurs in frame-level thresholding.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250378845A1Systems and methods for soft event detection with event-level thresholding

Publication Date: 2025.12.11 MITSUBISHI ELECTRIC RESEARCH LABORATORIES INC

AI summary

A system for event detection in time-series data comprises a memory configured to store computer-executable instructions and one or more processors configured to execute the instructions to process the time-series data to make a hard decision on a time span of an event indicative of continuous activity of the event within the time-series data and make a soft decision on a presence of the event for the entire time span. The one or more processors are further configured to apply an event-level threshold to the soft decision on the presence of the event for the entire time span to produce a result of the event detection and output the result of the event detection.