Acoustic Event Detection via Subpart Sequence Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing acoustic event detection systems require time-consuming and expensive annotation of training data and often lack accuracy in detection and false alarm rates, especially in IoT and smart home applications.
Innovation Solution
The system models acoustic events as sequences of subparts using unsupervised subspace clustering techniques, allowing for the training of classifiers that identify and score these subparts without requiring explicit event labeling, and employs sequence decoding to recognize the proper temporal sequence of subparts for improved detection performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing recognition systems are trained on acoustic training data sequences with labeled events, then the system can detect acoustic events, but the training process becomes time-consuming and expensive
Solution Approach 1:
The patent segments acoustic events into multiple subparts (onset, development, offset) with distinct acoustic characteristics. By training the classifier to recognize these subparts separately rather than requiring complete event labeling, the system reduces annotation effort while maintaining detection accuracy. The sequence decoder then reconstructs complete events from detected subparts.
2Reliability
If existing recognition systems use traditional event-level training, then the system structure is simple, but the detection accuracy and false alarm rate are insufficient
Solution Approach 1:
The system divides event detection into subpart detection (by classifier) and sequence reconstruction (by sequence decoder). This segmentation improves detection rate by capturing partial events that traditional methods miss, while the modular architecture manages complexity through clear separation of concerns.
Solution Approach 2:
The patent adds a temporal sequence dimension to detection by modeling events as ordered sequences of subparts. The sequence decoder operates in this temporal dimension to reconstruct complete events, improving detection accuracy without requiring complete initial labeling.
3Reliability
If the system models events at coarse granularity, then the training is simpler, but the discriminative power and detection reliability are reduced
Solution Approach 1:
By segmenting events into distinct acoustic subparts (onset, development, offset) with different characteristics, the system achieves fine-grained modeling that improves discriminative power and reduces false alarms. Each subpart can be detected independently with higher confidence.
Solution Approach 2:
The system applies different detection criteria and scoring thresholds to different subparts based on their local acoustic characteristics. This local quality approach allows optimized detection for each subpart type, improving overall reliability while managing complexity through localized processing.
Data Source
AI summary
Techniques are provided for acoustic event detection. A methodology implementing the techniques according to an embodiment includes extracting acoustic features from a received audio signal. The acoustic features may include, for example, one or more short-term Fourier transform frames, or other spectral energy characteristics, of the audio signal. The method also includes applying a trained classifier to the extracted acoustic features to identify and label acoustic event subparts of the audio signal and to generate scores associated with the subparts. The method further includes performing sequence decoding of the acoustic event subparts and associated scores to detect target acoustic events of interest based on the scores and temporal ordering sequence of the event subparts. The classifier is trained on acoustic event subparts that are generated through unsupervised subspace clustering techniques applied to training data that includes target acoustic events.