Acoustic Scene Classification via Event Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing acoustic scene classification techniques face challenges in identifying and generalizing sound events in real environments due to the need for manual definition and selection of specific events, which becomes impractical with the large number of events and their variable occurrence across environments.

Innovation Solution

The approach identifies generic event types by partitioning audio signals into short-event, long-event, and background frames based on change measures, allowing for improved feature extraction and classification without requiring specific event detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual definition and selection of specific sound events is performed, then event identification accuracy is improved, but device complexity and ease of operation deteriorate due to the large number of events and variable occurrence across environments

Engineering Contradiction:
Improveevent identification accuracyVSAvoidcomplexity of event definition and selection
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system automatically defines and selects sound events by analyzing acoustic patterns in audio recordings without requiring manual intervention. The processor identifies events based on detected acoustic patterns and automatically adds them to the event list, allowing the system to serve itself in the event definition task.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system changes the approach from manual event definition to automated pattern-based event detection. By detecting acoustic patterns and deriving events from these patterns, the system transforms the parameter of event identification from human-defined categories to algorithmically-derived events based on acoustic characteristics.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If manual definition and selection of specific sound events is performed, then event identification accuracy is improved, but ease of operation worsens due to the impracticality of managing large number of events

Engineering Contradiction:
Improveevent identification accuracyVSAvoidease of event management
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The system automatically defines and selects sound events by analyzing acoustic patterns in audio recordings without requiring manual intervention. The processor identifies events based on detected acoustic patterns and automatically adds them to the event list, allowing the system to serve itself in the event definition task.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If specific event detection is implemented, then classification accuracy for known events is improved, but adaptability to new environments and unseen events deteriorates

Engineering Contradiction:
Improveclassification accuracyVSAvoidadaptability to new environments
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system uses acoustic pattern detection as a universal mechanism that can identify both known and unknown events across different environments. By detecting acoustic patterns rather than relying on pre-defined event categories, the system achieves multi-functionality in event identification that adapts to various acoustic scenes.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system changes the approach from manual event definition to automated pattern-based event detection. By detecting acoustic patterns and deriving events from these patterns, the system transforms the parameter of event identification from human-defined categories to algorithmically-derived events based on acoustic characteristics.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3701528B1Segmentation-based feature extraction for acoustic scene classification
Publication Date: 2023.03.15 HUAWEI TECH CO LTD
  • EP3701528B1 patent drawingFigure 1
  • EP3701528B1 patent drawingFigure 2
  • EP3701528B1 patent drawingFigure 3

AI summary

An apparatus and a method for acoustic scene classification of a block of audio samples. The block is partitioned into frames in the time domain. For each frame of a plurality of frames of the block, a change measure between the frame and a preceding frame of the block is calculated. The frame is assigned to one of a set of short-event frames, a set of long-event frames, and a set of background frames, based on the calculated change measure. The feature vector is determined based on a feature computed from the set of short-event frames, the set of long-event frames, and the set of background frames.