A Method for Automatic Segmentation and Parameter Estimation of Bird Vocalizations

Hagai Barmatz, Dana Klein, Yoni Vortman, Sivan Toledo, Yizhar Lavner

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Animal vocalizations are ubiquitous produced by various taxa and represented in all habitats. Tracking and quantifying animal vocalizations is a basic necessity in various biological disciplines such as nature conservation and biomonitoring. With the advancement of digital recording technology, a huge amount of audio recordings is accumulated. Since manual annotation and an analysis of relevant acoustic features is impractical, development of reliable algorithms for an automatic analysis of birdsong is highly required. One of the first challenges in a birdsong analysis is that of segmentation of the acoustic signal, i.e. detection and demarcation of its basic elements or syllables prior to a further analysis. In this study, we present two simple unsupervised algorithms for automatic birdsong segmentation and parameter estimation. The algorithms are based on a smoothed envelope of the short-time energy of the signal, parameters derived from the fundamental frequency and short-time Fourier transform (STFT). The methods were evaluated using a small database of trill vocalizations recorded with high background noise. The algorithms output was compared to manual segmentation carried out by a human expert and to ground truth values obtained by using synthetic signals after which it was realized that they produced highly similar results. In summary, the methods are shown to accurately segment birdsong signals with high background noise levels. Since they are simple to implement, they could be of great benefit to bioacoustics researchers.

Original languageEnglish
Title of host publicationProceedings of IWSSIP 2019 - 2019 International Conference on Systems, Signals and Image Processing
EditorsSnjezana Rimac-Drlje, Drago Zagar, Irena Galic, Goran Martinovic, Denis Vranjes, Marija Habijan
PublisherIEEE Computer Society
Pages211-216
Number of pages6
ISBN (Electronic)9781728132273
DOIs
StatePublished - Jun 2019
Event26th International Conference on Systems, Signals and Image Processing, IWSSIP 2019 - Osijek, Croatia
Duration: 5 Jun 20197 Jun 2019

Publication series

NameInternational Conference on Systems, Signals, and Image Processing
Volume2019-June

Conference

Conference26th International Conference on Systems, Signals and Image Processing, IWSSIP 2019
Country/TerritoryCroatia
CityOsijek
Period5/06/197/06/19

Keywords

  • audio segmentation
  • audio signal processing
  • bioacoustics
  • bird vocalization
  • birdsong analysis

All Science Journal Classification (ASJC) codes

  • Software
  • Artificial Intelligence
  • Signal Processing
  • Computer Networks and Communications

Fingerprint

Dive into the research topics of 'A Method for Automatic Segmentation and Parameter Estimation of Bird Vocalizations'. Together they form a unique fingerprint.

Cite this