DI-UMONS : Dépôt institutionnel de l’université de Mons

Recherche transversale
Rechercher
(titres de publication, de périodique et noms de colloque inclus)
2016-08-28 - Article/Dans un journal avec peer-review - Anglais - 10 page(s)

Florentin Juliette , Dutoit Thierry , Verlinden Olivier , "Identification of European woodpecker species in audio recordings from their drumming rolls" in Ecological Informatics, 35, 61-70, http://dx.doi.org/10.1016/j.ecoinf.2016.08.006

  • Edition : Elsevier, Amsterdam (The Netherlands)
  • Codes CREF : Sciences de l'ingénieur (DI2000), Intelligence artificielle (DI1180), Ornithologie (DI3166), Acoustique (DI1264)
  • Unités de recherche UMONS : Mécanique rationnelle, Dynamique et Vibrations (F703)
  • Instituts UMONS : Institut NUMEDIART pour les Technologies des Arts Numériques (Numédiart)
Texte intégral :

Abstract(s) :

(Anglais) Drumming sounds are substantial clues when searching audio recordings for the presence of woodpeckers. Woodpeckers use drumming for territory defence and mate attraction to such an extent that some species have no vocalisations for these functions. This implies that drumming bears species markers. This hypothesis stands at the root of our project to develop an autonomous program for the identification of drumming species. To proceed, we assembled a database of 361 recordings from open-access bird sound archives. The recordings were for nine drumming species found on the European continent. Focusing on the signal below 1500 Hz, we reviewed all audio files and extracted 2665 drumming rolls. For recordings where multiple birds were present, the drumming rolls were attributed to individual birds. This allowed keeping track of the time interval between successive rolls. The characteristic traits of drumming are decidedly temporal. Consequently, the spectral features that have been successful in other recent bird identification studies were not applicable in our case. We developed specialized drumming parameters and automated their calculation.We then performed a t-SNE dimensionality reduction to visualise the dataset and to demonstrate that our parameters detached the different classes properly. Eventually, a k-NN algorithm accurately labelled 87.2% of the submitted test samples. The time structure within the drumming rolls (speed, acceleration) provided the critical features. The duration of the rolls followed in importance. The results compare well to existing literature and attest to the feasibility of monitoring European woodpecker species by tracking drumming.

Identifiants :
  • DOI : 10.1016/j.ecoinf.2016.08.006