Publication detail

Feature comparison under different noise conditions for gunshot detection task

HRABINA, M. SIGMUND, M.

Original Title

Feature comparison under different noise conditions for gunshot detection task

Type

conference paper

Language

English

Original Abstract

Presented work investigates performance of three feature sets (LPC, LPCC and MFCC) in distinguishing gunshots from non-gunshot, mostly urban sounds under white noise conditions ranging from 30 dB to 0 dB with 10 dB step, including clean signal. Results show, that LPC coefficients are best at 30 dB with comparable results achieved by LPCC. MFCC are significantly better than others at 20 dB and 10 dB. Performance at 0 dB was balanced between LPC and MFCC – LPC had more true detections and MFCC achieved better score for false alarms.

Keywords

gunshot detection, feature analysis, linear predictive coding coefficients, cepstrum, noise

Authors

HRABINA, M.; SIGMUND, M.

Released

28. 8. 2017

ISBN

978-80-214-5526-9

Book

Proceedings of IEEE Student Branch Conference Mikulov 2017

Edition

1

Pages from

29

Pages to

33

Pages count

4

BibTex

@inproceedings{BUT138779,
  author="Martin {Hrabina} and Milan {Sigmund}",
  title="Feature comparison under different noise conditions for gunshot detection task",
  booktitle="Proceedings of IEEE Student Branch Conference Mikulov 2017",
  year="2017",
  series="1",
  pages="29--33",
  isbn="978-80-214-5526-9"
}