2025_programme: Automated Classification of Vocalisations from Wild and Captive Seal Populations

Back


  • Day: June 16, Monday
      Location / Time: D. CHLOE at 15:10-15:30
  • Last minutes changes: -
  • Session: 08b. Bioacoustics and Soundscape
    Organiser(s): Jennifer Miksis-Olds, Giacomo Giorli
    Chairperson(s): Jennifer Miksis-Olds, Giacomo Giorli
  • Lecture: Automated Classification of Vocalisations from Wild and Captive Seal Populations
    Paper ID: 2300
    Author(s): William Butler, Harrison Smith, Wil Lloyd, Marios Impraimakis, Andrew Barnes, Alan Hunter
    Presenter: William Butler
    Abstract: Seals are known to regularly make a variety of vocalisations with no current standard for categorising these calls these into different types. Currently, most proposed methods are based on qualitative and subjective measures with varying granularity and quantity of classes between the different methods. Understanding the differences and similarities of call category density and overall vocalisation frequency between wild and captive seals is critically important for the study of seal behaviour and the viability of collecting training data based on captive seal calls for research and training of ML-based seal detection models. However, not much is currently known about the behavioural differences between captive and wild seals. In this paper, two new classes of seal vocalisations are presented and empirically evaluated using a fine-tuned ResNet model, trained on pre-processed spectrogram representations of the calls. We expand Nowak’s S1, S2 and S3 with newly proposed “SP4” and “SP5” seal vocalisation classes, while also finding that the S3 class is acoustically heterogeneous and may contain several functionally different vocalisation types.
      Download the full paper
    This paper is a candidate for the "Prof. Leif Bjørnø Best Student Paper Award (for students under 35)"
  • Corresponding author: Mr William Butler
    Affiliation: University of Bath
    Country: United Kingdom

Back