Large Acoustic Datasets Research Articles

A post‐processing framework for assessing BirdNET identification accuracy and community composition

Passively collected acoustic data have become increasingly common in wildlife research and have prompted the development of machine‐learning approaches to extract and classify large sets of audio files. BirdNET is an open‐source automatic prediction model that is popular because of its lack of training requirements for end users. Several studies have sought to test the accuracy of BirdNET and illustrate its potential in occupancy modelling of single or multiple species. However, these techniques either require extensive statistical knowledge or computational power to be applied to large datasets. In addition, there is a lack of comparisons of occupancy and community composition calculated using BirdNET and typical field methods. Here we develop a framework for assessing the accuracy of BirdNET using generalized linear mixed models to determine species‐specific confidence score thresholds. We then compare community composition under our model and another post‐processing approach to field data collected from co‐located point count surveys in northeastern Vermont. Our framework outperformed the other post‐processing method and resulted in species composition similar to that of point count surveys. Our work highlights the potential mismatch between accuracy and confidence score and the importance of developing species‐specific thresholds. The framework can facilitate research on large acoustic datasets and can be applied to output from BirdNET or other automatic prediction models.

Automated detection of Bornean white-bearded gibbon (Hylobates albibarbis) vocalizations using an open-source framework for deep learning.

Passive acoustic monitoring is a promising tool for monitoring at-risk populations of vocal species, yet, extracting relevant information from large acoustic datasets can be time-consuming, creating a bottleneck at the point of analysis. To address this, an open-source framework for deep learning in bioacoustics to automatically detect Bornean white-bearded gibbon (Hylobates albibarbis) "great call" vocalizations in a long-term acoustic dataset from a rainforest location in Borneo is adapted. The steps involved in developing this solution are described, including collecting audio recordings, developing training and testing datasets, training neural network models, and evaluating model performance. The best model performed at a satisfactory level (F score = 0.87), identifying 98% of the highest-quality calls from 90 h of manually annotated audio recordings and greatly reduced analysis times when compared to a human observer. No significant difference was found in the temporal distribution of great call detections between the manual annotations and the model's output. Future work should seek to apply this model to long-term acoustic datasets to understand spatiotemporal variations in H. albibarbis' calling activity. Overall, a roadmap is presented for applying deep learning to identify the vocalizations of species of interest, which can be adapted for monitoring other endangered vocalizing species.

Potential of K-means clustering for preliminary labeling of acoustic data samples

Labeling data for use in supervised learning algorithms can be a long and arduous process, especially for large datasets. This work explores if an unsupervised k-means clustering algorithm can act as a tool to efficiently obtain preliminary labels for large acoustic datasets. This approach is tested on vector sensor data from Monterey Accelerated Research System (MARS). To determine the optimal number of clusters in the k-means algorithm, the silhouette method is used. The clustering of different types of input data will be compared. Specifically, time average spectra from the pressure sensor, the acoustic particle velocity, and intensity will be considered. In addition, the results of clustering data from different times of day and months over a two-year period will be compared. Results of the clustering will be presented along with an assessment of the potential for using a k-means clustering as a preliminary labeling tool for large acoustic sets. [Undergraduate research funded by the College of Physical and Mathematical Sciences as Brigham Young University.]

Characteristics and spatiotemporal variation of sei whale (Balaenoptera borealis) downsweeps recorded in Atlantic Canada.

The call characteristics and vocal behaviour of sei whales (Balaenoptera borealis) off eastern Canada, including potential spatiotemporal variation, is poorly understood. Such information can improve the performance of automated detector-classifiers, enhancing the accuracy and efficiency of identifying sei whales in large acoustic datasets. Ultimately, these data can be used to understand the occurrence, distribution, and population structure of sei whales in Atlantic Canada. We measured sei whale downsweep characteristics recorded from six locations off Nova Scotia (NS) and Newfoundland and Labrador (NFLD), Canada over a two-year period (2015-2017), and examined variation between call subtypes (singlets, doublets, triplets+), and seasons (Spring, Summer, Fall, Winter). We found that downsweeps had a mean duration of 1.58 s, sweeping from 75.66 to 34.22 Hz, with a peak frequency of 43.89 Hz and an intra-call interval of 2.22 s. Most call characteristics did not vary between location, subtype, or season; however, significantly longer downsweeps occurred off NS, within doublet calls, and in fall months. We also found that NFLD had a higher proportion of doublets (70%) than NS (52%). This variation may be evidence of acoustically diverging sei whale populations, as well as useful for improving detector-classifiers of sei whales in the region.

Improving deep learning acoustic classifiers with contextual information for wildlife monitoring

Bioacoustics, the exploration of animal vocalizations and natural soundscapes, has emerged as a valuable tool for studying species within their habitats, particularly those that are challenging to observe. This approach has broadened the horizons of biodiversity assessment and ecological research. However, monitoring wildlife with acoustic recorders produces large volumes of data that can be labor-intensive to analyze. Deep learning has recently transformed many computational disciplines by enabling the automated processing of large and complex datasets and has gained attention within the bioacoustics community. Despite the revolutionary impact of deep learning on acoustic detection and classification, attaining both high detection accuracy and low false positive rates in bioacoustics remains a significant challenge. An intriguing yet unexplored avenue for enhancing deep learning in bioacoustics involves the utilization of contextual information, such as time and location, to discern animal vocalizations within acoustic recordings. As a first case study, a multi-branch Convolutional Neural Network (CNN) was developed to classify 22 different bird songs using spectrograms as a first input, and spatial metadata as a secondary input. A comparison was made to a baseline model with only spectrogram input. A geographical prior neural network was trained, separately, to estimate the probability of a species occurring at a given location. The output of this network was combined with the baseline CNN. As a second case study, temporal data and spectrograms were used as input to a multi-branch CNN for the detection of Hainan gibbon (Nomascus hainanus) calls, the world’s rarest primate. Our findings demonstrate that adding metadata to the bird song classifier significantly improves classification performance, with the highest improvement achieved using the geographical prior model (F1-score of 87.78% compared to 61.02% for the baseline model). The multi-branch CNNs also proved efficient (F1-scores of 76.87% and 78.77%) and simpler to use than the geographical prior. In the second case study, our findings revealed a decrease in false positives by 63% (94% of the calls were detected) when the metadata was used by the multi-branch CNN, and an increase of 19% in gibbon detection. This study has uncovered an exciting new avenue for improving classifier performance in bioacoustics. The methodology described in this study can assist ecologists, wildlife management teams, and researchers in reducing the amount of time spent analyzing large acoustic datasets obtained from passive acoustic monitoring studies. Our approach can be adapted and applied to other calling species, and thus tailored to other use cases.

Artificial intelligence (BirdNET) supplements manual methods to maximize bird species richness from acoustic data sets generated from regional monitoring

Processing methods that maximize species richness from acoustic recordings obtained from regional monitoring programs can increase detections of uncommon, rare, and cryptic species and provide key information on species status and distribution. Using data from regional bird monitoring in Yukon, Canada, we (1) compared the number of bird species detected (species richness) and the cost associated with four acoustic processing methods ( Listening, Visual Scanning, Recognizer, and Recognizer with Validation) and (2) combined Listening and Recognizer with Validation information to increase detections of all bird species at the ecoregion scale. We used comprehensive Visual Scanning to detect all bird species on the recordings. We processed ∼1% of the recordings using Listening and detected 56% of the bird community with 71.5 h of human effort. We used Recognizer (multispecies recognizer BirdNET) with Validation and detected 89% of the bird community with ∼22% of the effort required for Visual Scanning (56 and 257 h, respectively). As an application of our approach, we combined Listening and Recognizer with Validation to process recordings from five northern ecoregions and found a 23%–63% increase in the number of bird species detected with little additional effort. Combining Listening and Recognizer with Validation can maximize species detections from large passive acoustic monitoring data sets.

Adapting deep learning models to new acoustic environments - A case study on the North Atlantic right whale upcall

Passive acoustic monitoring is increasingly being used for studying marine mammals, leading to the accumulation of large acoustic datasets. Analyzing these datasets becomes impractical without automated detection and classification software. Detectors and classifiers based on deep neural networks have shown great potential, but their performance is often limited by the availability of sufficient quantities of annotated training samples, and their application restricted to the specific acoustic environment(s) from which their training data were collected. We address these limitations by employing transfer learning, a deep learning concept whereby knowledge from a source domain is transferred to a target domain. Specifically, we considered two different underwater acoustic environments as the source and target domains. The objective was to use a deep neural network that had been trained in one environment with abundant annotated training samples, and optimize its performance in the other environment where the annotated training samples were limited. Training and testing were conducted using three acoustic datasets containing North Atlantic right whale (Eubalaena glacialis) upcalls. Experiments show that adapting a trained model to the new environment led to a substantial improvement in recall from 70% to 85%, while maintaining a low false-positive rate of less than 5 per hour. The methodology is implemented as an open-source Python tool to facilitate the creation of more tailored deep learning-based acoustic detectors and classifiers for North Atlantic right whale vocalizations and other stereotyped marine mammal calls.

Manual Versus Semiautomated Bioacoustic Analysis Methods of Multiple Vocalizations in Tricolored Blackbird Colonies

Abstract Bioacoustic monitoring can reveal aspects of animal behavior because many species vocalize in association with certain behaviors. Despite this, bioacoustics remain infrequently used to monitor animal behavior because of lack of knowledge of how vocalizations relate to behavior and the challenge of efficiently analyzing the large acoustic data sets necessary to capture relevant behaviors. Vocalizations and associated behaviors have been previously established for the colonial tricolored blackbird Agelaius tricolor, but efficient analysis of the acoustic data remains a challenge. Previous work with tricolored blackbird acoustic data relied on manually listening to recordings, which is not practical on large scales. Using software to automatically detect vocalizations of interest has potential to reduce analysis time. However, automated detection is prone to errors often caused by faint vocalizations, overlapping calls, and background noise. Thus, incorporating components of manual and automated analysis of acoustic data sets remains essential. To address these challenges, we deployed autonomous recording units at three tricolored blackbird colonies in California from 2019 to 2021 and analyzed acoustic data using a manual and a semiautomated analysis method. Specifically, we used tricolored blackbird male song, male chorus, female song, hatchling call, nestling call, and fledgling call to determine the approximate timing of breeding stages and number of breeding attempts, or pulses, for each colony. We found that using a semiautomated approach was more time efficient than manual analysis, while using comparable numbers of recordings and obtaining equivalent information from the colonies. The odds of correct detections of vocalizations using the semiautomated method were generally lower for fainter vocalizations and colonies with high background noise. Overall, the semiautomated approach had tolerable rates of recall, precision, false positives, and false negatives. Our methodology adds to a growing body of literature addressing acoustic analyses, especially for colonial species and where questions of breeding phenology are important.

Passive acoustic exploration of the ocean

Encoded in the spectral density, spatial variability, and directionality (spatial coherence) of the ambient sound field is information on the generation mechanisms of sound and the properties of the ocean propagation environment and its boundaries. Through field and observatory measurements, and analytical and computational models of the underwater sound field, a research program has been pursued that asks, “What can we learn about the ocean by listening?” Large acoustic data sets have been exploited to develop an estimate of the effective source level per unit area of surface generated noise. In complement, a methodology for precisely partitioning the sound field into ship generated and wind generated components by exploiting the vertical noise directionality has been demonstrated. Models of the spatial properties of wind driven and ship generated sound have further been used to estimate the geoacoustic properties of the seabed, the depth of mix layer, depth-averaged pH, and localize a source in bearing and azimuth using only a pair of vertically oriented omnidirectional hydrophones. An autonomous passive acoustic profiler, The Deep Acoustic Lander (DAL), recently made measurements of the ambient sound field from the surface to the bottom of the Challenger Deep, Mariana Trench, precisely determining the mixing of a locally- and distantly generated contributions to the sound field. Meanwhile, DAL measurements at the Endeavour hydrothermal vent field have revealed components of the sound field generated by vent activity.

Silbido profundo: An open source package for the use of deep learning to detect odontocete whistles.

This work presents an open-source matlab software package for exploiting recent advances in extracting tonal signals from large acoustic data sets. A whistle extraction algorithm published by Li, Liu, Palmer, Fleishman, Gillespie, Nosal, Shiu, Klinck, Cholewiak, Helble, and Roch [(2020). Proceedings of the International Joint Conference on Neural Networks, July 19-24, Glasgow, Scotland, p. 10] is incorporated into silbido, an established software package for extraction of cetacean tonal calls. The precision and recall of the new system were over 96% and nearly 80%, respectively, when applied to a whistle extraction task on a challenging two-species subset of a conference-benchmark data set. A second data set was examined to assess whether the algorithm generalized to data that were collected across different recording devices and locations. These data included 487 h of weakly labeled, towed array data collected in the Pacific Ocean on two National Oceanographic and Atmospheric Administration (NOAA) cruises. Labels for these data consisted of regions of toothed whale presence for at least 15 species that were based on visual and acoustic observations and not limited to whistles. Although the lack of per whistle-level annotations prevented measurement of precision and recall, there was strong concurrence of automatic detections and the NOAA annotations, suggesting that the algorithm generalizes well to new data.

A Robust Method to Automatically Detect Fin Whale Acoustic Presence in Large and Diverse Passive Acoustic Datasets

The growing availability of long-term and large-scale passive acoustic recordings open the possibility of monitoring the vocal activity of elusive oceanic species, such as fin whales (Balaenoptera physalus), in order to acquire knowledge on their distribution, behavior, population structure and abundance. Fin whales produce low-frequency and high-intensity pulses, both as single vocalizations and as song sequences (only males) which can be detected over large distances. Numerous distant fin whales producing these pulses generate a so-called chorus, by spectrally and temporally overlapping single vocalizations. Both fin whale pulses and fin whale chorus provide a distinct source of information on fin whales present at different distances to the recording location. The manual review of vast amounts of passive acoustic data for the presence of single vocalizations and chorus by human experts is, however, time-consuming, often suffers from low reproducibility and in its entirety, it is practically impossible. In this publication, we present and compare robust algorithms for the automatic detection of fin whale choruses and pulses which yield good performance results (i.e., false positive rates < 3% and true positive rates > 76%) when applied to real-world passive acoustic datasets characterized by vast amounts of data, with only a small proportion of the data containing the target sounds, and diverse soundscapes from the Southern Ocean.

Low-cost open-source recorders and ready-to-use machine learning approaches provide effective monitoring of threatened species

Passive acoustic monitoring is a powerful tool for monitoring vocally active taxa. Automated signal recognition software reduces the expert time needed for recording analyses and allows researchers and managers to manage large acoustic datasets. The application of state-of-the-art techniques for automated identification, such as Convolutional Neural Networks, may be challenging for ecologists and managers without informatics or engineering expertise. Here, we evaluated the use of AudioMoth — a low-cost and open-source sound recorder — to monitor a threatened and patchily distributed species, the Eurasian bittern (Botaurus stellaris). Passive acoustic monitoring was carried out across 17 potential wetlands in north Spain. We also assessed the performance of BirdNET — an automated and freely available classifier able to identify over 3000 bird species — and Kaleidoscope Pro — a user-friendly recognition software — to detect the vocalizations and the presence of the target species. The percentage of presences and vocalizations of the Eurasian bittern automatically detected by BirdNET and Kaleidoscope Pro software was compared to manual annotations of 205 recordings. The species was effectively recorded up to distances of 801–900 m, with at least 50% of the vocalizations uttered within that distance being manually detected; this distance was reduced to 601–700 m when considering the analyses carried out using Kaleidoscope Pro. BirdNET detected the species in 59 of the 63 (93.7%) recordings with known presence of the species, while Kaleidoscope detected the bittern in 62 recordings (98.4%). At the vocalization level, BirdNet and Kaleidoscope Pro were able to detect between 76 and 78%, respectively, of the vocalizations detected by a human observer. Our study highlights the ability of AudioMoth for detecting the bittern at large distances, which increases the potential of that technique for monitoring the species at large spatial scales. According to our results, a single AudioMoth could be useful for monitoring the species' presence in wetlands of up to 150 ha. Our study proves the utility of passive acoustic monitoring, coupled with BirdNET or Kaleidoscope Pro, as an accurate, repeatable, and cost-efficient method for monitoring the Eurasian bittern at large spatial and temporal scales. Nonetheless, further research should evaluate the performance of BirdNET on a larger number of species, and under different recording conditions (e.g., more closed habitats), to improve our knowledge about BirdNET's ability to perform bird monitoring. Future studies should also aim to develop an adequate protocol to perform effective passive acoustic monitoring of the Eurasian bittern.

A new method employing species‐specific thresholding identifies acoustically overlapping bats

AbstractPassive acoustic detectors are increasingly used for monitoring biodiversity, particularly for echolocating bat species (Microchiroptera). However, identification of calls collected at large scales is hindered by substantial variation within and between species, and the considerable time investment needed to manually identify acoustic data. We use acoustic data from 14 species of echolocating bats, occurring in temperate forests and woodlands of southeastern Australia to build a supervised classification model that identifies species from large acoustic datasets. Acoustic data from hand‐release (39,567) and free‐flying (8851) bat calls were used to build a predictive model, which was then validated using field‐collected calls (149,097) from the same region. We maximized the model fit per species by validating the associated confidence scores against manually identified presence and absence values. This allowed us to model the identification success of each species as a function of the confidence score. From this relationship, we set specific thresholds for accepting species identification, enabling more accurate classification of calls and identification of multiple bat species within a single acoustic recording. Including calls from manually identified free‐flying bats improved overall identification accuracy, including a 60% improvement for bats that navigate in open spaces. Assigning species‐specific thresholds achieved substantial improvements in overall model confidence, with functionally meaningful changes in the identification of species exhibiting considerable acoustic overlap in time and frequency measures. Research into the ecological requirements of species is hampered by problems with identification. Our research illustrates that internal train–test validation overestimates model accuracy particularly for species that were in low abundance or for uncommon species, which are acoustically similar to more common ones. Recognizing this, we set specific thresholds per species below which identifications were not accepted. Our method is particularly relevant in locations with high overlap in species' call parameters, which can result in false negatives in preference for species that are easier to identify because of the common practice of assigning one species per acoustic recording. This research proposes a cautious method to substantially reduce the burden of manual identification of large acoustic datasets.

Automatic detection and classification of baleen and toothed whale calls via machine learning approaches over instantaneous wide areas in the Gulf of Maine received on a coherent hydrophone array

Large acoustic data sets are typically generated from ocean observations with a 160-element coherent hydrophone array and correspondingly larger volumes of acoustic detection events stem from coherent array processing. Beamforming enhances detection signal to-noise ratio, significantly improving detection ranges, as well as providing signal bearing. Here, we develop and train algorithms for the automatic detection and classification of baleen and toothed whale calls present in multiple beamformed spectrograms spanning 360 degree azimuths generated via the passive ocean acoustic waveguide remote sensing technique in the following six categories for the Gulf of Maine: Fin, Sei, Minke, Humpback, unidentified baleen whale downsweep chirps, and general toothed whale encompassing echolocation clicks and whistles below 4 kHz. The classifiers include random forest, support vector machine (SVM), and decision tree applied to hand-engineered features, as well as Convolutional Neural Network (CNN)-based model on the per-channel energy normalization transform (PCEN) applied directly to beamformed spectrogram imagery. Total accuracy of 95% and average F1-score of 85% are achieved using random forest classifier. The processing flow, including beamforming, PCEN extraction and call classification, run in real-time making the methods suitable for real-world applications, such as marine mammal monitoring and mitigation in ocean hydrocarbon prospecting and wind farm installations.

Detecting and reducing heterogeneity of error in acoustic classification

Abstract Passive acoustic monitoring can be an effective method for monitoring species, allowing the assembly of large audio datasets, removing logistical constraints in data collection and reducing anthropogenic monitoring disturbances. However, the analysis of large acoustic datasets is challenging and fully automated machine learning processes are rarely developed or implemented in ecological field studies. One of the greatest uncertainties hindering the development of these methods is spatial generalisability—can an algorithm trained on data from one place be used elsewhere? We demonstrate that heterogeneity of error across space is a problem that could go undetected using common classification accuracy metrics. Second, we develop a method to assess the extent of heterogeneity of error in a random forest classification model for six Amazonian bird species. Finally, we propose two complementary ways to reduce heterogeneity of error, by (i) accounting for it in the thresholding process and (ii) using a secondary classifier that uses contextual data. We found that using a thresholding approach that accounted for heterogeneity of precision error reduced the coefficient of variation of the precision score from a mean of 0.61 ± 0.17 (SD) to 0.41 ± 0.25 in comparison to the initial classification with threshold selection based on F‐score. The use of a secondary, contextual classification with thresholding selection accounting for heterogeneity of precision reduced it further still, to 0.16 ± 0.13, and was significantly lower than the initial classification in all but one species. Mean average precision scores increased, from 0.66 ± 0.4 for the initial classification, to 0.95 ± 0.19, a significant improvement for all species. We recommend assessing—and if necessary correcting for—heterogeneity of precision error when using automated classification on acoustic data to quantify species presence as a function of an environmental, spatial or temporal predictor variable.

Echolocation click discrimination for three killer whale ecotypes in the Northeastern Pacific.

Three killer whale ecotypes are found in the Northeastern Pacific: residents, transients, and offshores. These ecotypes can be discriminated in passive acoustic data based on distinct pulsed call repertoires. Killer whale acoustic encounters for which ecotypes were assigned based on pulsed call matching were used to characterize the ecotype-specific echolocation clicks. Recordings were made using seafloor-mounted sensors at shallow (∼120 m) and deep (∼1400 m) monitoring locations off the coast of Washington state. All ecotypes' echolocation clicks were characterized by energy peaks between 12 and 19 kHz, however, resident clicks featured sub peaks at 13.7 and 18.8 kHz, while offshore clicks had a single peak at 14.3 kHz. Transient clicks were rare and were characterized by lower peak frequencies (12.8 kHz). Modal inter-click intervals (ICIs) were consistent but indistinguishable for resident and offshore killer whale encounters at the shallow site (0.21-0.22 s). Offshore ICIs were longer and more variable at the deep site, and no modal ICI was apparent for the transient ecotype. Resident and offshore killer whale ecotype may be identified and distinguished in large passive acoustic datasets based on properties of their echolocation clicks, however, transient echolocation may be unsuitable in isolation as a cue for monitoring applications.

Development of deep neural networks for marine mammal call detection using an open-source, user friendly tool

As the collection of large acoustic datasets used to monitor marine mammals increases, so too does the need for expedited and reliable detection of accurately classified bioacoustic signals. Deep learning methods of detection and classification are increasingly proposed as a means of addressing this processing need. These image recognition and classification methods include the use of a neural networks that independently determine important features of bioacoustic signals from spectrograms. Recent marine mammal call detection studies report consistent performance even when used with datasets that were not included in the network training. We present here the use of DeepSqueak, a novel open-source tool originally developed to detect and classify ultrasonic vocalizations from rodents in a low-noise, laboratory setting. We have trained networks in DeepSqueak to detect marine mammal vocalizations in comparatively noisy, natural acoustic environments. DeepSqueak utilizes a regional convolutional neural network architecture within an intuitive graphical user interface that provides automated detection results independent of acoustician expertise. Using passive acoustic data from two hydrophones on the Ocean Observatories Initiative’s Coastal Endurance Array, we developed networks for humpback whales, delphinids, and fin whales. We report performance and limitations for use of this detection method for each species.

PAMGuard: Open-source detection, classification, and Localization software

The PAMGuard open-source software project was established in 2006 to provide a toolbox and app for the detection, classification, and localization (DCL) of bioacoustic signals. Releases typically attract between 1500 and 3500 downloads. Primarily developed for marine mammals, users can configure PAMGuard to work with a wide variety of hardware configurations (single hydrophones, to complex 3D arrays) and DCL algorithms, depending on the species of interest. A Java code base and integrated multi-threading allow multiple DCL algorithms to be run in parallel enabling rapid analysis of large acoustic datasets. A comprehensive data management system and a suite of interactive displays then allow users to visualize and annotate resulting data at temporal scales from milliseconds to years. This combination of advanced analysis algorithms and powerful data visualization in both real-time, or reviewing data post-hoc, means PAMGuard provides both automatic and a proven and reliable human-in-the-loop approach to analysis. Developers can use the program’s infrastructure to rapidly develop and deploy new algorithms; recently, an international collaboration has improved support for terrestrial species and developed a new module to import acoustic deep learning models, enabling users to take advantage of PAMGuard’s data management and visualization capabilities alongside the latest automated acoustic classifiers.

Caller ID for Risso\u2019s and Pacific White-sided dolphins

Tracking species with expanding ranges is crucial to conservation efforts and some typically temperate marine species are spreading northward into the Arctic Ocean. Risso’s (Gg) and Pacific white-sided (Lo) dolphins have been documented spreading poleward. Further, they make very similar sounds, so it is difficult for both human analysts and classification algorithms to tell them apart. Using automatic detectors and classifiers on large acoustic datasets would improve the efficiency of monitoring these species. variational mode decomposition (VMD) provides both an easier visualization tool for human analysts and exhibited robustness to background noise while extracting features in pulsed signals with very similar spectral properties. The goal of this work was to develop a new visualization tool using VMD and a statistics-based classification algorithm to differentiate similar pulsed signals. The proposed VMD method achieved 81% accuracy, even when using audio files with low SNR that did not have concurrent visual survey data. While many dolphins whistle, pulsed signals are one of the more useful vocalizations to use in detection and classification because of their species-specific acoustic features. Automating the VMD method and expanding it to other dolphin species that have very similar pulsed signals would complement current detection and classification methods and lead to a more complete understanding of ecosystem dynamics under a changing climate.

Using a Novel Visualization Tool for Rapid Survey of Long-Duration Acoustic Recordings for Ecological Studies of Frog Chorusing

Continuous recording of environmental sounds could allow long-term monitoring of vocal wildlife, and scaling of ecological studies to large temporal and spatial scales. However, such opportunities are currently limited by constraints in the analysis of large acoustic data sets. Computational methods and automation of call detection require specialist expertise and are time consuming to develop, therefore most biological researchers continue to use manual listening and inspection of spectrograms to analyze their sound recordings. False-color spectrograms were recently developed as a tool to allow visualization of long-duration sound recordings, intending to aid ecologists in navigating their audio data and detecting species of interest. This paper explores the efficacy of using this visualization method to identify multiple frog species in a large set of continuous sound recordings and gather data on the chorusing activity of the frog community. We found that, after a phase of training of the observer, frog choruses could be visually identified to species with high accuracy. We present a method to analyze such data, including a simple R routine to interactively select short segments on the false-color spectrogram for rapid manual checking of visually identified sounds. We propose these methods could fruitfully be applied to large acoustic data sets to analyze calling patterns in other chorusing species.