Analytical approaches for evaluating passive acoustic monitoring data: A case study of avian vocalizations

Evaluating community-wide temporal sampling in passive acoustic monitoring: A comprehensive study of avian vocal patterns in subtropical montane forests

Background

From passive acoustic monitoring (PAM) recordings, the vocal activity rate (VAR), vocalizations per unit of time, can be calculated and is essential for assessing bird population abundance. However, VAR is subject to influences from a range of factors, including species and environmental conditions. Identifying the optimal sampling design to obtain representative acoustic data for VAR estimation is crucial for research objectives. PAM commonly uses temporal sampling strategies to decrease the volume of recordings and the resources needed for audio data management. Yet, the comprehensive impact of this sampling approach on VAR estimation remains insufficiently explored.

Methods

In this study, we used vocalizations extracted from recordings of 12 bird species, taken at 14 PAM stations situated in subtropical montane forests over a four-month period, to assess the impact of temporal sampling on VAR across three distinct scales: short-term periodic, diel, and hourly. For short-term periodic sampling analysis, we employed hierarchical clustering analysis (HCA) and the coefficient of variation (CV). Generalized additive models (GAMs) were utilized for diel sampling analysis, and we determined the average difference in VAR values per minute for the hourly sampling analysis.

Results

We identified significant day and species-specific VAR fluctuations. The survey season was divided into five segments; the earliest two showed high variability and are best avoided for surveys. Data from days with heavy rain and strong winds showed reduced VAR values and should be excluded from analysis. Continuous recordings spanning at least seven days, extending to 14 days is optimal for minimizing sampling variance. Morning chorus recordings effectively capture the majority of bird vocalizations, and hourly sampling with frequent, shorter intervals aligns closely with continuous recording outcomes.

Conclusions

While our findings are context-specific, they highlight the significance of strategic sampling in avian monitoring, optimizing resource utilization and enhancing the breadth of monitoring efforts.

Global birdsong embeddings enable superior transfer learning for bioacoustic classification

Automated bioacoustic analysis aids understanding and protection of both marine and terrestrial animals and their habitats across extensive spatiotemporal scales, and typically involves analyzing vast collections of acoustic data. With the advent of deep learning models, classification of important signals from these datasets has markedly improved. These models power critical data analyses for research and decision-making in biodiversity monitoring, animal behaviour studies, and natural resource management. However, deep learning models are often data-hungry and require a significant amount of labeled training data to perform well. While sufficient training data is available for certain taxonomic groups (e.g., common bird species), many classes (such as rare and endangered species, many non-bird taxa, and call-type) lack enough data to train a robust model from scratch. This study investigates the utility of feature embeddings extracted from audio classification models to identify bioacoustic classes other than the ones these models were originally trained on. We evaluate models on diverse datasets, including different bird calls and dialect types, bat calls, marine mammals calls, and amphibians calls. The embeddings extracted from the models trained on bird vocalization data consistently allowed higher quality classification than the embeddings trained on general audio datasets. The results of this study indicate that high-quality feature embeddings from large-scale acoustic bird classifiers can be harnessed for few-shot transfer learning, enabling the learning of new classes from a limited quantity of training data. Our findings reveal the potential for efficient analyses of novel bioacoustic tasks, even in scenarios where available training data is limited to a few samples.

ANIMAL-SPOT enables animal-independent signal detection and classification using deep learning

Bioacoustic research spans a wide range of biological questions and applications, relying on identification of target species or smaller acoustic units, such as distinct call types. However, manually identifying the signal of interest is time-intensive, error-prone, and becomes unfeasible with large data volumes. Therefore, machine-driven algorithms are increasingly applied to various bioacoustic signal identification challenges. Nevertheless, biologists still have major difficulties trying to transfer existing animal- and/or scenario-related machine learning approaches to their specific animal datasets and scientific questions. This study presents an animal-independent, open-source deep learning framework, along with a detailed user guide. Three signal identification tasks, commonly encountered in bioacoustics research, were investigated: (1) target signal vs. background noise detection, (2) species classification, and (3) call type categorization. ANIMAL-SPOT successfully segmented human-annotated target signals in data volumes representing 10 distinct animal species and 1 additional genus, resulting in a mean test accuracy of 97.9%, together with an average area under the ROC curve (AUC) of 95.9%, when predicting on unseen recordings. Moreover, an average segmentation accuracy and F1-score of 95.4% was achieved on the publicly available BirdVox-Full-Night data corpus. In addition, multi-class species and call type classification resulted in 96.6% and 92.7% accuracy on unseen test data, as well as 95.2% and 88.4% regarding previous animal-specific machine-based detection excerpts. Furthermore, an Unweighted Average Recall (UAR) of 89.3% outperformed the multi-species classification baseline system of the ComParE 2021 Primate Sub-Challenge. Besides animal independence, ANIMAL-SPOT does not rely on expert knowledge or special computing resources, thereby making deep-learning-based bioacoustic signal identification accessible to a broad audience.

Refining manual annotation effort of acoustic data to estimate bird species richness and composition: The role of duration, intensity, and time

Manually annotating audio files for bird species richness estimation or machine learning validation is a time-intensive task. A premium is placed on the subselection of files that will maximize the efficiency of unique additional species identified, to be used for future analyses. Using acoustic data collected in 17 plots, we created 60 subsetting scenarios across three gradients: intensity (minutes in an hour), day phase (dawn, morning, or both), and duration (number of days) for manual annotation. We analyzed the effect of these variables on observed bird species richness and assemblage composition at both the local and entire study area scale. For reference, results were also compared to richness and composition estimated by the traditional point count method. Intensity, day phase, and duration all affected observed richness in decreasing respective order. These variables also significantly affected observed assemblage composition (in the same order of effect size), but only the day phase produced compositional dissimilarity that was due to phenological traits of individual bird species, rather than differences in species richness. All annotation scenarios requiring equal sampling effort to point counts yielded higher species richness than the point count method. Our results show that a great majority of species can be obtained by annotating files at high sampling intensities (every 3 or 6 min) in the morning period (post-dawn) over a duration of two days. Depending on a study's aim, different subsetting parameters will produce different assemblage compositions, potentially omitting rare or crepuscular species, species representing additional functional groups and natural history guilds, or species of higher conservation concern. We do not recommend one particular subsetting regime for all research objectives, but rather present multiple scenarios for researchers to understand how intensity, day phase, and duration interact to identify the best subsetting regime for one's particular research interests.