Anomaly detection for environmental noise monitoring

Octave-band sound pressure level is the preferred measure for continuous environmental noise monitoring over raw audio because accepted standards and devices exist, these data do not compromise voice privacy, and thus an octave-band sound meter can legally collect data in public. By setting up an experiment that continuously monitors octave-band sound pressure level in a residential street, we show daily noise-level patterns correlated to human activities. Directly applying well-known anomaly detection algorithms including one-class support vector machine, replicator neural network, and principal component analysis based anomaly detection shows low performance in the collected data because these standard algorithms are unable to exploit the daily patterns. Therefore, principal component analysis anomaly detection with time-varying mean and the covariance matrix over each hour, is proposed in order to detect abnormal acoustic events in the octave band measurements of the residential-noise-monitoring application. The proposed method performs at 0.83 in recall, 0.88 in precision and 0.85 in F-measure on the evaluation data set.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2018

Author dc:creator

• Phan, Duc H.

Contributors dc:contributor

• Jones, Douglas L.

• Anomaly detection, environmental noise monitoring

Statement dc:rights

• copyright 2018 Duc Phan

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/101116

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/101116