Exploratory analysis of yellow mongoose vocalization: detection from in-the-wild recordings and call classification
| Type of publication: | Conference paper |
| Citation: | Hovsepyan_INTERSPEECH_2026_2026 |
| Booktitle: | Proceedings of Interspeech |
| Year: | 2026 |
| Abstract: | Understanding the vocal repertoire of social animals, such as the yellow mongoose (YM), is crucial for deciphering their social structure and evolutionary history. In this study, we present an exploratory analysis of YM vocalization detection and classification using both signal processing and machine learning approaches. The analysis uses two distinct datasets: (1) expert-annotated, clean pup vocalizations (n=940), and (2) noisy field recordings (n=29) captured with directional microphones. Our results indicate that handcrafted features, originally developed for human speech modeling to represent syllables, are effective for animal vocalization classification, suggesting potential shared acoustic structures. Although detecting YM vocalizations in wild recordings proved challenging, our objective is to aid annotators by flagging potential vocalization segments. The results presented in this work provide a foundation for scalable, semi-automated analysis of animal vocal repertoires. |
| Main Research Program: | Human-AI Teaming |
| Additional Research Programs: |
AI for Everyone |
| Keywords: | animal VAD, animal vocalizations, handcrafted features, language evolution, yellow mongoose |
| Projects: |
EVOLANG Idiap |
| Authors: | |
| Added by: | [UNK] |
| Total mark: | 0 |
|
Attachments
|
|
|
|
|
Notes
|
|
|
|
|
