Saved in:
Bibliographic Details
Main Authors: Sarkar, Eklavya, -Doss, Mathew Magimai.
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.10190
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908650471161856
author Sarkar, Eklavya
-Doss, Mathew Magimai.
author_facet Sarkar, Eklavya
-Doss, Mathew Magimai.
contents Animal vocalizations contain sequential structures that carry important communicative information, yet most computational bioacoustics studies average the extracted frame-level features across the temporal axis, discarding the order of the sub-units within a vocalization. This paper investigates whether discrete acoustic token sequences, derived through vector quantization and gumbel-softmax vector quantization of extracted self-supervised speech model representations can effectively capture and leverage temporal information. To that end, pairwise distance analysis of token sequences generated from HuBERT embeddings shows that they can discriminate call-types and callers across four bioacoustics datasets. Sequence classification experiments using $k$-Nearest Neighbour with Levenshtein distance show that the vector-quantized token sequences yield reasonable call-type and caller classification performances, and hold promise as alternative feature representations towards leveraging sequential information in animal vocalizations.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10190
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Towards Leveraging Sequential Structure in Animal Vocalizations
Sarkar, Eklavya
-Doss, Mathew Magimai.
Machine Learning
Animal vocalizations contain sequential structures that carry important communicative information, yet most computational bioacoustics studies average the extracted frame-level features across the temporal axis, discarding the order of the sub-units within a vocalization. This paper investigates whether discrete acoustic token sequences, derived through vector quantization and gumbel-softmax vector quantization of extracted self-supervised speech model representations can effectively capture and leverage temporal information. To that end, pairwise distance analysis of token sequences generated from HuBERT embeddings shows that they can discriminate call-types and callers across four bioacoustics datasets. Sequence classification experiments using $k$-Nearest Neighbour with Levenshtein distance show that the vector-quantized token sequences yield reasonable call-type and caller classification performances, and hold promise as alternative feature representations towards leveraging sequential information in animal vocalizations.
title Towards Leveraging Sequential Structure in Animal Vocalizations
topic Machine Learning
url https://arxiv.org/abs/2511.10190