Prepoznava oglašanja nočnih sov

Majkič, Tara (author), Marolt, Matija (mentor)

Dolgoročno spremljanje oglašanja nočnih sov omogoča globlji vpogled v njihovo vedenje in populacijske trende, kar pomembno prispeva k razumevanju biotske raznovrstnosti in ekološkega stanja Slovenije. Ročno pregledovanje zvočnih posnetkov, zajetih v daljšem časovnem obdobju, je časovno zamuden postopek pri analizi populacij nočnih sov. Diplomska naloga obravnava ta problem in ga rešuje z avtomatizacijo procesa, pri čemer uporablja obstoječe globoke nevronske modele, ki so bili prvotno naučeni na velikih zbirkah ptičjih oglašanj, nato pa prilagojeni na lastnih zvočnih posnetkih nočnih sov, ki prebivajo v Sloveniji. Model, ki ni omogočal prilagoditve, je bil uporabljen za pridobivanje vektorskih vložitev, ki so bile uporabljene kot podatki za učenje in testiranje enostavnih klasifikatorjev na osnovi metod strojnega učenja. Dosežki diplomske naloge bodo v podporo Nacionalnemu inštitutu za biologijo pri analizi populacij nočnih sov v Sloveniji.

• Language: Slovenian
• Keywords: avtomatska analiza zvoka, nočne sove, bioakustika, globoko učenje, vektorske vložitve, strojno učenje
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2026
• PID: 20.500.12556/RUL-179420
• COBISS.SI-ID: 270205955
• Publication date in RUL: 13.02.2026

Secondary language

• Language: English
• Title: Identification of nocturnal owl calls

Abstract

Long-term monitoring of nocturnal owl vocalizations offers deeper insight into their behavior and population trends, significantly enhancing the understanding of biodiversity and ecological state of Slovenia. Manually inspecting audio recordings collected over a longer period of time is a time-consuming task when analyzing nocturnal owl populations. This diploma thesis addresses the issue by automating the process with existing deep neural network models that were originally trained on large collections of bird vocalizations and later adapted to locally recorded owl sounds from Slovenia. The model that did not support adaptation was used to extract embeddings, which were then used as input data for training and testing simple classifiers based on traditional machine learning methods. The results of this diploma thesis will support the National Institute of Biology in the analysis of nocturnal owl populations in Slovenia.

• Keywords: automatic sound analysis, nocturnal owls, bioacoustics, deep learning, embeddings, machine learning