Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NBM: an Open Dataset for the Acoustic Monitoring of Nocturnal Migratory Birds in Europe

Louis Airale, Adrien Pajot, Juliette Linossier

TL;DR

The paper tackles the lack of fine-grained nocturnal bird vocalization data in Europe by introducing the NBM dataset, a publicly available collection with 13,359 time-and-frequency annotated vocalizations from 117 Western Palearctic species. It demonstrates the utility of precise localization with a two-stage spectrogram-based object detector, built on an EfficientNetV2 backbone and enhanced with self-attention and sinusoidal positional encodings, achieving a mean average precision of 0.67 on a 45-species test subset. Despite training on a smaller dataset, the detector remains competitive with BirdNet in multi-label evaluation, highlighting the value of high-quality annotations over sheer data volume. The work emphasizes open science and provides data and code to support scalable, fine-grained acoustic monitoring for conservation of nocturnal migrants in Europe, with future plans to expand species coverage and dataset diversity.

Abstract

The persisting threats on migratory bird populations highlight the urgent need for effective monitoring techniques that could assist in their conservation. Among these, passive acoustic monitoring is an essential tool, particularly for nocturnal migratory species that are difficult to track otherwise. This work presents the Nocturnal Bird Migration (NBM) dataset, a collection of 13,359 annotated vocalizations from 117 species of the Western Palearctic. The dataset includes precise time and frequency annotations, gathered by dozens of bird enthusiasts across France, enabling novel downstream acoustic analysis. In particular, we prove the utility of this database by training an original two-stage deep object detection model tailored for the processing of audio data. While allowing the precise localization of bird calls in spectrograms, this model shows competitive accuracy on the 45 main species of the dataset with state-of-the-art systems trained on much larger audio collections. These results highlight the interest of fostering similar open-science initiatives to acquire costly but valuable fine-grained annotations of audio files. All data and code are made openly available.

NBM: an Open Dataset for the Acoustic Monitoring of Nocturnal Migratory Birds in Europe

TL;DR

The paper tackles the lack of fine-grained nocturnal bird vocalization data in Europe by introducing the NBM dataset, a publicly available collection with 13,359 time-and-frequency annotated vocalizations from 117 Western Palearctic species. It demonstrates the utility of precise localization with a two-stage spectrogram-based object detector, built on an EfficientNetV2 backbone and enhanced with self-attention and sinusoidal positional encodings, achieving a mean average precision of 0.67 on a 45-species test subset. Despite training on a smaller dataset, the detector remains competitive with BirdNet in multi-label evaluation, highlighting the value of high-quality annotations over sheer data volume. The work emphasizes open science and provides data and code to support scalable, fine-grained acoustic monitoring for conservation of nocturnal migrants in Europe, with future plans to expand species coverage and dataset diversity.

Abstract

The persisting threats on migratory bird populations highlight the urgent need for effective monitoring techniques that could assist in their conservation. Among these, passive acoustic monitoring is an essential tool, particularly for nocturnal migratory species that are difficult to track otherwise. This work presents the Nocturnal Bird Migration (NBM) dataset, a collection of 13,359 annotated vocalizations from 117 species of the Western Palearctic. The dataset includes precise time and frequency annotations, gathered by dozens of bird enthusiasts across France, enabling novel downstream acoustic analysis. In particular, we prove the utility of this database by training an original two-stage deep object detection model tailored for the processing of audio data. While allowing the precise localization of bird calls in spectrograms, this model shows competitive accuracy on the 45 main species of the dataset with state-of-the-art systems trained on much larger audio collections. These results highlight the interest of fostering similar open-science initiatives to acquire costly but valuable fine-grained annotations of audio files. All data and code are made openly available.

Paper Structure

This paper contains 13 sections, 1 equation, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. The NBM Dataset
  4. Data collection and Annotation
  5. Database properties
  6. NBM Test Set
  7. The NBM Detection Model: an Object Detection Baseline
  8. Model Architecture
  9. Model Evaluation
  10. Performances on the Detection Task
  11. Comparison with BirdNet
  12. Discussion & Limitations
  13. Conclusion

Figures (4)

  • Figure 1: (a) Data volume breakdown between the original database and the XC extra base. (b) Annotation repartition between the original and XC bases for the 20 most common species. We refer the reader to \ref{['tab:sp_key_name']} for the correspondence between short and proper Latin names.
  • Figure 2: The object detection approach allows a fine-grained time and frequency localization of short flight calls.
  • Figure 3: Overview of the presented NBM bird sound detection modular architecture.
  • Figure 4: Learned posterior probabilities for call frequencies vs data distribution on three species: dunlin (left), tree pipit (middle), and song thrush (right). Note the alignment between ground truth and predicted frequencies, learned solely through the use of the positional encoding.