Title:
ABNORMALITY LEVEL ESTIMATION DEVICE, ABNORMALITY LEVEL ESTIMATION METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2021/019672
Kind Code:
A1
Abstract:
The present invention prevents erroneous detections in which a normal sound is erroneously determined to be abnormal. A registered normal sound detection device 2 calculates the abnormality level of an observation signal. A feature value extraction unit 24 extracts a feature value of a fixed length from a time series acoustic signal having an arbitrarily defined length. An abnormality level calculation unit 25 corrects the abnormality level calculated from the observation signal such that the value of the abnormality level decreases as the similarity between the observation signal and a registered normal sound increases. The abnormality level calculation unit 25 calculates said similarity via a similarity function learned by using the feature value extracted by the feature value extraction unit 24 from the time series acoustic signal, which includes at least a normal sound.
Inventors:
KOIZUMI YUMA (JP)
SAITO SHOICHIRO (JP)
SAITO SHOICHIRO (JP)
Application Number:
PCT/JP2019/029777
Publication Date:
February 04, 2021
Filing Date:
July 30, 2019
Export Citation:
Assignee:
NIPPON TELEGRAPH & TELEPHONE (JP)
International Classes:
G01H17/00; G01M99/00
Domestic Patent References:
| WO2019049688A1 | 2019-03-14 | |||
| WO2018042616A1 | 2018-03-08 |
Foreign References:
| JP2018022014A | 2018-02-08 | |||
| JPH09166483A | 1997-06-24 | |||
| JPS58100734A | 1983-06-15 | |||
| JP2017090606A | 2017-05-25 | |||
| JP2013253847A | 2013-12-19 | |||
| CN106323452A | 2017-01-11 |
Other References:
KOIZUMI, YUMA ET AL.: "SNIPER_Few-shot learning for anomaly detection to minimize false-negative rate with ensured true-positive rate", ICASSP 2019 - 2019 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), 16 April 2019 (2019-04-16), pages 915 - 919, XP033566221, Retrieved from the Internet [retrieved on 20191016], DOI: 10.1109/ICASSP.2019.8683667
Attorney, Agent or Firm:
NAKAO, Naoki et al. (JP)
Download PDF: