Bad: Taming the Bioacoustic Data Deluge with a Bat Activity Detector

Stefano Ciapponi; Santiago Martinez Balvanera; Andrea Cesaretti; Elisabetta Farella; Kate E. Jones

cs.SDSep 29, 2026

Bad: Taming the Bioacoustic Data Deluge with a Bat Activity Detector

Authors: Stefano Ciapponi, Santiago Martinez Balvanera, Andrea Cesaretti, Elisabetta Farella, Kate E. Jones

Organizations: Fondazione Bruno Kessler · University of Trento · University College London

Abstract

Passive Acoustic Monitoring of bats generates massive ultrasonic datasets (>27 GB/night per node), straining edge storage and battery life. Legacy triggers fail against acoustic confusers, while deep models exceed microcontroller limits. We present a hardware-aware Bat Activity Detector (BAD) specifically designed to discriminate bat calls from hard biological and environmental confusers across variable sampling rates (192-384 kHz). Tailored for the Silicon Labs EFM32PG26 (MVP) in 8-bit integer precision, our model achieves 100 percent hardware offload across all 14 layers (17.2 KB Flash, 73.1 KB RAM). End-to-end preprocessing (74.00 ms for 76 frames) and inference (30.00 ms) of 100 ms clips at 192 kHz require 104.00 ms per clip. On spatially out-of-domain recordings under a realistic low-prevalence regime (r_pos = 0.05), BAD achieves an AUC-ROC of 0.9748 and suppresses 99.4% of non-target noise frames while retaining 65.3% of bat calls - delivering a >33x precision gain over classical Goertzel baselines.

Figures & tables

Figure 1 : Spectrogram of a bat echolocation call sequence recorded at 256 kHz across a 0.5 s duration, illustrating rapid frequency-modulated (FM) sweeps within the 10–120 kHz band.

Figure 2 : Bat vs Non-bat 200 ms clips resampled at 256Khz from Section 3 , after preprocessing highlighted in Section 4 .

SR (kHz)	Segment (ms)	Epoch	Accuracy	Precision	Recall	Specificity	FPR	F1 Score	AUC-ROC	Avg Precision
192	50	9	0.7720	0.8908	0.6200	0.9240	0.0760	0.7311	0.8076	0.8587
256	50	2	0.7410	0.9003	0.5420	0.9400	0.0600	0.6767	0.7484	0.8192
384	50	5	0.7590	0.8453	0.6340	0.8840	0.1160	0.7246	0.7998	0.8538
192	100	5	0.9160	0.9298	0.9000	0.9320	0.0680	0.9146	0.9599	0.9684
256	100	3	0.9030	0.9296	0.8720	0.9340	0.0660	0.8999	0.9495	0.9572
384	100	16	0.9070	0.9111	0.9020	0.9120	0.0880	0.9065	0.9532	0.9626

Table 1 : Bat vs. Non-Bat Classification Validation Performance Across Sample Rates and Segment Lengths (Overall Best Metrics in Bold ).

Balanced Prevalence ( $r_{\text{pos}}=0.5024$ )
Method	AUC	AP	Prec	Sens	Spec	FPR	F1	$\mathbf{F_{0.5}}$
AMP	0.3915	0.4158	53.3%	86.1%	23.8%	76.2%	65.8%	57.7%
Goertzel	0.6066	0.5845	57.1%	80.3%	39.0%	61.1%	66.7%	60.6%
BAD (Ours)	0.9732	0.9743	95.7%	87.2%	96.1%	3.9%	91.3%	93.9%
Realistic Low Prevalence ( $r_{\text{pos}}=0.0513$ )
AMP	0.3849	0.0385	5.7%	85.0%	24.0%	76.0%	10.7%	7.0%

Table 2 : Performance comparison on

100\text{~ms}

clips under

F_{0.5}

-optimized decision thresholds (

N=10{,}000

, uk_diff split).

Metric	Windowing	Cropping	Resizing	SMS	EMA	FFT
Per Frame ( $\mu\text{s}$ )	$52.63$	$65.79$	$39.47$	$39.47$	$184.21$	$592.11$
76 Frames ( ms )	$4.00$	$5.00$	$3.00$	$3.00$	$14.00$	$45.00$
Metric	Total Prep ( ms )		Inference ( ms )		Total Latency ( ms )
Duration	$74.00$		$30.00$		$104.00$

Table 3 : Stage Execution Times and Overall Processing Latency (

100\text{~ms}

Audio Clips,

192\text{~kHz}

76\text{~Frames}

80\text{~MHz}

Explore similar work

SEABAD: A Tropical Bird Activity Detection Dataset for Passive Acoustic Monitoring
May 20, 2026Muhammad Mun'im Ahmad Zabidi, Mohd Yamani Idna Idris, Norisma IdrisPassive Acoustic Monitoring Biodiversity Monitoring
Faculty of Computer Science and Information Technology, Universiti Malaya, Kuala Lumpur, Malaysia
Smart Passive Acoustic Monitoring: Embedding a Classifier on AudioMoth Microcontroller
May 5, 2026Louis Lerbourg, Paul Peyret, Juliette Linossier +1Passive Acoustic Monitoring Cortex
Univ. Grenoble Alpes, CEA, List, Grenoble, France · Biophonia, France
Ultra-Compact CNN Architectures for Tropical Bird Audio Detection on Microcontrollers
Jul 22, 2026Muhammad Mun'im Ahmad Zabidi, Mohd Yamani Idna Idris, Norisma IdrisCortex Mobilenetv2
Faculty of Computer Science and Information Technology, Universiti Malaya, Kuala Lumpur, Malaysia

May 20, 2026, cs.SDJ/K move · Enter open · S save

SEABAD: A Tropical Bird Activity Detection Dataset for Passive Acoustic Monitoring

Muhammad Mun'im Ahmad Zabidi, Mohd Yamani Idna Idris, Norisma Idris

Faculty of Computer Science and Information Technology, Universiti Malaya, Kuala Lumpur, Malaysia

Passive Acoustic Monitoring Biodiversity Monitoring Df-Wild Dataset Thai

Open paper AskPDF arXiv

Passive acoustic monitoring (PAM) enables large-scale biodiversity assessment, but continuous recording generates large amounts of non-informative audio, creating challenges for storage, power consumption, and long-term edge deployment. Bird audio detection (BAD), which identifies bird vocalizations, can reduce this burden by filtering irrelevant recordings before downstream analysis. However, most BAD systems are trained on temperate datasets despite tropical soundscapes being denser, more species-rich, and acoustically unpredictable. To address this gap, we introduce SEABAD (Southeast Asian Bird Activity Detection), a dataset of 50,000 curated three-second clips from Southeast Asian soundscapes, evenly balanced between bird-present and bird-absent samples. The dataset spans 1,677 bird species and is standardized to 16 kHz mono audio for embedded and low-power inference. We developed a dual-branch curation pipeline: a six-stage positive-label workflow applied to Xeno-Canto recordings, alongside six source-specific negative-label extractions from environmental datasets. These procedures reduced class imbalance by 13.7% (Gini coefficient: 0.601 to 0.519). A manual audit of 1,000 positive clips confirmed 97.8% +/- 0.9% labeling accuracy. Baseline experiments using MobileNetV3-Small achieved 99.57% +/- 0.25% accuracy and 0.9985 +/- 0.0002 AUC across three random seeds. SEABAD and the full curation pipeline are publicly released to support tropical BAD research and energy-efficient acoustic monitoring.

Abstract

Explore similar work

SEABAD: A Tropical Bird Activity Detection Dataset for Passive Acoustic Monitoring

Smart Passive Acoustic Monitoring: Embedding a Classifier on AudioMoth Microcontroller

Ultra-Compact CNN Architectures for Tropical Bird Audio Detection on Microcontrollers