Acoustic Abnormality Estimation for Rare Normal Sound Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional unsupervised anomaly detection methods often result in false positives, incorrectly labeling normal data that appears less frequently as anomalous due to the use of negative logarithmic likelihood.

Innovation Solution

An abnormality estimation device that optimizes an anomaly estimation model to minimize the difference between the anomaly degrees of normal data appearing more frequently and less frequently, using a larger weight for less frequent normal data and minimizing the average weighted anomaly degrees.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If negative logarithmic likelihood is used for anomaly detection, then the anomaly degree can be calculated based on generative probability, but normal data appearing less frequently is incorrectly determined as anomalous (false positive)

Engineering Contradiction:

Improveanomaly degree calculation accuracyVSAvoidfalse positive rate

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent changes the optimization parameter from minimizing average anomaly degree to minimizing the difference between anomaly degrees of frequent and infrequent normal data. This parameter change resolves the contradiction by adjusting what the model optimizes for, thereby reducing false positives while maintaining anomaly detection capability

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

Instead of directly minimizing the anomaly degree of infrequent normal data (which would increase false negatives), the patent inverts the approach by minimizing the difference between anomaly degrees of frequent and infrequent normal data. This indirect approach reduces false positives without compromising the detection of actual anomalies

Inventive Principle:

Principle #13The other way round (Inversion)

2Ease of manufacture

If the model is optimized to minimize average anomaly degree of all normal data, then training is simplified, but the anomaly degree of infrequent normal data becomes excessively high

Engineering Contradiction:

Improvemodel training simplicityVSAvoidanomaly degree estimation accuracy

Core Design Contradiction:

Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent modifies the optimization parameter from average anomaly degree to the difference between anomaly degrees of frequent and infrequent normal data. This parameter change maintains training simplicity while improving anomaly degree estimation accuracy for infrequent normal data

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12387743B2Abnormality estimation device, abnormality estimation method, and program

Publication Date: 2025.08.12 NT T INC

AI summary

Provided is an abnormality estimation device capable of appropriately determining normal data appearing less frequently as normal. The abnormality estimation device includes an estimation unit that estimates an anomaly degree of an acoustic signal, by using an abnormality estimation model that is optimized while using a set of normal sounds and is optimized so as to minimize a difference between an anomaly degree of a normal sound appearing more frequently and an anomaly degree of a normal sound appearing less frequently.