Wigner–Ville-Based Conditional Autoencoder For Biosonar Generation And Joint Range And Off-Axis Estimation With A Single Hydrophone
Abstract
Estimating the range and off-axis angle of the sperm whale biosonar using a single passive hydrophone is a longstanding bioacoustics problem and a prerequisite for unbiased spectral, inter-pulse interval, and behavioral analyses. We propose a Conditional AutoEncoder (CAE) conditioned via Feature-wise Linear Modulation (FiLM), which applies label-derived affine transformations at every decoder resolution to jointly reconstruct time-frequency representations and regress source range and off-axis angle. Several input configurations combining STFT, Wigner–Ville derived cross-representations (MWVD), and complex-valued Wideband Ambiguity Function (WAF) are compared under a two-phase training strategy. Joint reconstruction- and-regression followed by label-conditioned decoder refinement with a frozen image encoder. The best configuration achieves a RMSE of approximately 276 m in range and 14° in off-axis angle, outperforming existing single-hydrophone baselines. The model additionally supports conditional synthesis of physically consistent click representations, corroborating the bent-horn model of sperm whale sound production.
Related
- trFlow: ultrafast all-atom protein conformational ensemble generation via conditional flow matching
- OmniTCR: a foundation model unifying T cell receptor recognition prediction and conditional sequence generation
- HELICS: Biobank-scale Conditional Synthetic Genome Generation via Latent Flow Matching
- WhaleSonic: Recordings of cetacean vocal behaviour with hydrophone arrays in Northern Norway
- Audiovisual diarization of overlapping click trains in sperm whale (Physeter macrocephalus) vocal sparring using a three-hydrophone array
BibTeX
@misc{deloustal2026wignervillebased,
title = {{Wigner–Ville-Based Conditional Autoencoder For Biosonar Generation And Joint Range And Off-Axis Estimation With A Single Hydrophone}},
author = {Nicolas Deloustal and Hervé Glotin and Sébastien Paris},
howpublished = {HAL},
year = {2026},
month = oct,
url = {https://hal.science/hal-05723856v2},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Wigner–Ville-Based Conditional Autoencoder For Biosonar Generation And Joint Range And Off-Axis Estimation With A Single Hydrophone
AU - Nicolas Deloustal
AU - Hervé Glotin
AU - Sébastien Paris
PY - 2026
DA - 2026/10/16
PB - HAL
UR - https://hal.science/hal-05723856v2
AB - Estimating the range and off-axis angle of the sperm whale biosonar using a single passive hydrophone is a longstanding bioacoustics problem and a prerequisite for unbiased spectral, inter-pulse interval, and behavioral analyses. We propose a Conditional AutoEncoder (CAE) conditioned via Feature-wise Linear Modulation (FiLM), which applies label-derived affine transformations at every decoder resolution to jointly reconstruct time-frequency representations and regress source range and off-axis angle. Several input configurations combining STFT, Wigner–Ville derived cross-representations (MWVD), and complex-valued Wideband Ambiguity Function (WAF) are compared under a two-phase training strategy. Joint reconstruction- and-regression followed by label-conditioned decoder refinement with a frozen image encoder. The best configuration achieves a RMSE of approximately 276 m in range and 14° in off-axis angle, outperforming existing single-hydrophone baselines. The model additionally supports conditional synthesis of physically consistent click representations, corroborating the bent-horn model of sperm whale sound production.
ER -