ArcticSoundsNET: BirdNET embeddings facilitate improved bioacoustic classification of Arctic species
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
In recent years, deep learning has become a popular solution for processing large ecological monitoring datasets. This rise in use has resulted in global classification models for a variety of data types and taxa, such as BirdNET, which classifies vocalizations of more than 6000 avian species from acoustic data. These global models can be useful pre-trained models for transfer learning, allowing researchers to more easily develop classifiers specialized to their datasets. However, the development of such models hinges on the availability of comprehensive, high-quality training data, which can be difficult to acquire, produce, and use. We present a novel pipeline for creating training data from a large and unlabeled dataset with minimal human oversight. We used this pipeline and BirdNET as our base model to develop a transfer-learning-based model, ArcticSoundsNET, using acoustic monitoring data from 205 sites across Alaska's Arctic Coastal Plain . We compared performance of ArcticSoundsNET with that of BirdNET to evaluate the effectiveness of our pipeline and success of the new model. We found that the ability of ArcticSoundsNET to detect and classify avian vocalizations in our data greatly exceeded that of BirdNET (AUC ROC = 0.888 for ArcticSoundsNET, AUC ROC = 0.593 for BirdNET). Importantly, our method for developing a training dataset is widely applicable for ecologists who do not have large amounts of labeled data, facilitating the creation of task-specific classification models. Developing such models is an essential step in using large acoustic datasets to support ecological conservation of critical species and habitats.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it