Unsupervised blue whale call detection using multiple time-frequency features
In the context of bio-acoustic sciences, call detection is a critical task for understanding the behaviour of marine mammals such as the blue whale species (Balaeonoptera musculus) considered in this work. In this paper we present an approach to blue whale call detection from an unsupervised perspec...
| Published in: | 2017 CHILEAN Conference on Electrical, Electronics Engineering, Information and Communication Technologies (CHILECON) |
|---|---|
| Main Authors: | , , , , |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
IEEE
2017
|
| Subjects: | |
| Online Access: | https://doi.org/10.1109/CHILECON.2017.8229663 https://repositorio.uchile.cl/handle/2250/169122 |
| Summary: | In the context of bio-acoustic sciences, call detection is a critical task for understanding the behaviour of marine mammals such as the blue whale species (Balaeonoptera musculus) considered in this work. In this paper we present an approach to blue whale call detection from an unsupervised perspective. To achieve this, we use temporal and spectral features of audio acquired with a marine autonomous recording unit. The features considered are 46-dimensional and include the mel frequency ceptrum coefficients, chromagrams, and other scalar quantities; these features were then grouped via two different clustering algorithms. Our findings confirm the suitability of the proposed approach for isolating blue whale calls from other environmental sounds (as validated by a bio-acoustic specialist). This is a clear contribution for the annotation of blue whales calls, where the search for calls can now be performed by analysing the clusters identified instead of the entire recordings, thus saving time and effort for practitioners in bio-acoustics. |
|---|