Graph-Based Audio Classification Using Pre-Trained Models and Graph Neural Networks.

Sound classification plays a crucial role in enhancing the interpretation, analysis, and use of acoustic data, leading to a wide range of practical applications, of which environmental sound analysis is one of the most important. In this paper, we explore the representation of audio data as graphs in the context of sound classification. We propose a methodology that leverages pre-trained audio models to extract deep features from audio files, which are then employed as node information to build graphs. Subsequently, we train various graph neural networks (GNNs), specifically graph convolutional networks (GCNs), GraphSAGE, and graph attention networks (GATs), to solve multi-class audio classification problems. Our findings underscore the effectiveness of employing graphs to represent audio data. Moreover, they highlight the competitive performance of GNNs in sound classification endeavors, with the GAT model emerging as the top performer, achieving a mean accuracy of 83% in classifying environmental sounds and 91% in identifying the land cover of a site based on its audio recording. In conclusion, this study provides novel insights into the potential of graph representation learning techniques for analyzing audio data.

Soundscape Characterization Using Autoencoders and Unsupervised Learning.

Passive acoustic monitoring (PAM) through acoustic recorder units (ARUs) shows promise in detecting early landscape changes linked to functional and structural patterns, including species richness, acoustic diversity, community interactions, and human-induced threats. However, current approaches primarily rely on supervised methods, which require prior knowledge of collected datasets. This reliance poses challenges due to the large volumes of ARU data. In this work, we propose a non-supervised framework using autoencoders to extract soundscape features. We applied this framework to a dataset from Colombian landscapes captured by 31 audiomoth recorders. Our method generates clusters based on autoencoder features and represents cluster information with prototype spectrograms using centroid features and the decoder part of the neural network. Our analysis provides valuable insights into the distribution and temporal patterns of various sound compositions within the study area. By utilizing autoencoders, we identify significant soundscape patterns characterized by recurring and intense sound types across multiple frequency ranges. This comprehensive understanding of the study area's soundscape allows us to pinpoint crucial sound sources and gain deeper insights into its acoustic environment. Our results encourage further exploration of unsupervised algorithms in soundscape analysis as a promising alternative path for understanding and monitoring environmental changes.

Systematic review of machine learning methods applied to ecoacoustics and soundscape monitoring.

Soundscape ecology is a promising area that studies landscape patterns based on their acoustic composition. It focuses on the distribution of biotic and abiotic sounds at different frequencies of the landscape acoustic attribute and the relationship of said sounds with ecosystem health metrics and indicators (e.g., species richness, acoustic biodiversity, vectors of structural change, gradients of vegetation cover, landscape connectivity, and temporal and spatial characteristics). To conduct such studies, researchers analyze recordings from Acoustic Recording Units (ARUs). The increasing use of ARUs and their capacity to record hours of audio for months at a time have created a need for automatic processing methods to reduce time consumption, correlate variables implicit in the recordings, extract features, and characterize sound patterns related to landscape attributes. Consequently, traditional machine learning methods have been commonly used to process data on different characteristics of soundscapes, mainly the presence-absence of species. In addition, it has been employed for call segmentation, species identification, and sound source clustering. However, some authors highlight the importance of the new approaches that use unsupervised deep learning methods to improve the results and diversify the assessed attributes. In this paper, we present a systematic review of machine learning methods used in the field of ecoacoustics for data processing. It includes recent trends, such as semi-supervised and unsupervised deep learning methods. Moreover, it maintains the format found in the reviewed papers. First, we describe the ARUs employed in the papers analyzed, their configuration, and the study sites where the datasets were collected. Then, we provide an ecological justification that relates acoustic monitoring to landscape features. Subsequently, we explain the machine learning methods followed to assess various landscape attributes. The results show a trend towards label-free methods that can process the large volumes of data gathered in recent years. Finally, we discuss the need to adopt methods with a machine learning approach in other biological dimensions of landscapes.

Changes to an urban marina soundscape associated with COVID-19 lockdown in Guadeloupe.

In 2020, the COVID-19 pandemic led to government-enforced limits on activities worldwide, causing a marked reduction of human presence in outdoors environments, including in coastal areas that normally support substantial levels of boat traffic. These restrictions provided a unique opportunity to quantify the degree to which anthropogenic noise contributes to and impacts underwater soundscapes. In Guadeloupe, French West Indies, a significantly lower number of motor boats were recorded in the vicinity of the major urban marina during the peak of the first COVID-19 lockdown (April-May 2020), compared with the number recorded post-lockdown. The resumption of human activities at the end of May was correlated with a maximum increase of 6 decibels in the ambient noise level underwater. The change in noise level did not impact daily sound production patterns of vocal fishes, with increased activity at dusk seen both during and after the lockdown period. However, during the lockdown vocal activity was comprised of a reduced number of sounds, suggesting that anthropogenic noise has the potential to interfere with vocalization behaviours in fishes.

Changes on soundscapes reveal impacts of wildfires in the fauna of a Brazilian savanna.

Wildfire is a natural process in Brazilian savannas, but human activities alter fire regimes and threaten biodiversity. In this study, we used an ecoacoustics approach to assess fauna responses and recovery after wildfire in a Brazilian savanna. Six passive acoustic monitoring devices were used to record soundscapes before and after a wildfire a at burned and non-burned sites for one year and one month (September 2012 to September 2013). Power Spectral Density and the Acoustic Complexity Index were used to track biophony. Before the fire, the two sites had similar biophonic patterns (PSD: T = 1136, Z = 1.52, P = 0.12; ACI: T = 1117, Z = 1.10, P = 0.26) and soniferous species richness (Site 1 = 52 and Site 2 = 49). However, in the first two sessions of recordings after the fire, biophony became higher at the burned site during the day (PSD: T = 211 and 233; Z = 4.13 and 6.41; ACI: T = 120 and 469, Z = 5.14 and 7.07; all P < 0.00). During the night, biophony was usually higher at the non-burned site until May 2013 (PSD: T = 0 to 453; Z = 3.30 to 5.90; ACI: T = 333 to 491, Z = 3.80 to 4.93; all P < 0.00). Biophony became similar (P = 0.17 to 0.38) at the two sites or higher (P = 0.00 to 0.01) at the burned site from July to September 2013 (PSD: T = 55 to 1167; Z = 1.35 to 6.89; ACI: T = 719 to 1365, Z = 0.87 to 3.04). After the fire, a reduction of soniferous species at the burned site was observed for insects and bats. Both biophonic activity and soniferous species showed a tendency to recover one year after the fire, but there were still less species in September 2013 (non-burned = 43 and burned = 37) when compared to September 2012 at both sites (Site 1 = 52 and Site 2 = 49). Our results showed that changes in the natural regimes of fire can negatively impact the biodiversity and reinforce the need for monitoring protocols and inspection of wildfires.

Impacts of a large hydroelectric dam on the Madeira River (Brazil) on floodplain avifauna / Impacto de uma grande usina hidroelétrica sobre a avifauna de várzea doRio Madeira (Brasil)

Hydroelectric dams represent an important threat to seasonally flooded environments in the Amazon basin. We aimed to evaluate how a dam in the Madeira River, one of the largest tributaries of the Amazonas River, affected floodplain avifauna. Bird occurrence was recorded through simultaneous passive acoustic monitoring in early successional vegetation and floodplain forest downstream from the dam and upstream in sites impacted by permanent flooding after dam reservoir filling. Species were identified through manual inspection and semi-automated classification of the recordings. To assess the similarity in vegetation between downstream and upstream sites, we used Landsat TM/ETM+ composite images from before (2009-2011) and after (2016-2018) reservoir filling. Downstream and upstream floodplain forest sites were similar before, but not after dam construction. Early successional vegetation sites were already different before dam construction. We recorded 195 bird species. While species richness did not differ between upstream and downstream sites, species composition differed significantly. Ten species were indicators of early successional vegetation upstream, and four downstream. Ten species were indicators of floodplain forest upstream, and 31 downstream. Seven of 24 floodplain specialist species were detected by the semi-automated classification only upstream. While we found some bird species characteristic of early successional vegetation in the upstream sites, we did not find most species characteristic of tall floodplain forest. Predominantly carnivorous, insectivorous, and nectarivorous species appear to have been replaced by generalist and widely distributed species.

Barragens hidroelétricas representam uma importante ameaça a ambientes sazonalmente alagados na Amazônia. Avaliamos como uma barragem no Rio Madeira, um dos maiores tributários do Rio Amazonas, afetou a comunidade de aves de várzea. A ocorrência de aves foi registrada através de monitoramento acústico passivo simultâneo em vegetação em estágio sucessional inicial e floresta de várzea a jusante e em áreas a montante alagadas permanentemente após a formação do reservatório. Espécies foram identificadas por inspeção manual e classificação semi-automática das gravações. Para acessar a similaridade entre a vegetação a jusante e montante, utilizamos composições de imagens Landsat TM/ETM+ de antes (2009-2011) e após (2016-2018) a formação do reservatório. Sítios de floresta de várzea foram similares antes, mas não após o reservatório. Sítios de vegetação sucessional inicial já diferiam antes do reservatório. Registramos 195 espécies de aves. A riqueza de espécies não diferiu entre os sítios a jusante e montante, mas a composição de espécies diferiu significativamente. Dez espécies foram indicadoras de vegetação sucessional inicial a montante e quatro a jusante. Dez espécies foram indicadoras de floresta de várzea a montante e 31 a jusante. Sete de 24 espécies especialistas de várzea foram detectadas apenas a montante pelas classificações semi-automáticas. Encontramos algumas espécies típicas de vegetação sucessional inicial a montante, porém não encontramos a maioria de espécies típicas the floresta alta de várzea. Predominantemente, aves carnívoras, insetívoras e nectarívoras aparentam ter sido substituídas por espécies generalistas e amplamente distribuídas.

Impacts of a large hydroelectric dam on the Madeira River (Brazil) on floodplain avifauna

ABSTRACT Hydroelectric dams represent an important threat to seasonally flooded environments in the Amazon basin. We aimed to evaluate how a dam in the Madeira River, one of the largest tributaries of the Amazonas River, affected floodplain avifauna. Bird occurrence was recorded through simultaneous passive acoustic monitoring in early successional vegetation and floodplain forest downstream from the dam and upstream in sites impacted by permanent flooding after dam reservoir filling. Species were identified through manual inspection and semi-automated classification of the recordings. To assess the similarity in vegetation between downstream and upstream sites, we used Landsat TM/ETM+ composite images from before (2009-2011) and after (2016-2018) reservoir filling. Downstream and upstream floodplain forest sites were similar before, but not after dam construction. Early successional vegetation sites were already different before dam construction. We recorded 195 bird species. While species richness did not differ between upstream and downstream sites, species composition differed significantly. Ten species were indicators of early successional vegetation upstream, and four downstream. Ten species were indicators of floodplain forest upstream, and 31 downstream. Seven of 24 floodplain specialist species were detected by the semi-automated classification only upstream. While we found some bird species characteristic of early successional vegetation in the upstream sites, we did not find most species characteristic of tall floodplain forest. Predominantly carnivorous, insectivorous, and nectarivorous species appear to have been replaced by generalist and widely distributed species.

RESUMO Barragens hidroelétricas representam uma importante ameaça a ambientes sazonalmente alagados na Amazônia. Avaliamos como uma barragem no Rio Madeira, um dos maiores tributários do Rio Amazonas, afetou a comunidade de aves de várzea. A ocorrência de aves foi registrada através de monitoramento acústico passivo simultâneo em vegetação em estágio sucessional inicial e floresta de várzea a jusante e em áreas a montante alagadas permanentemente após a formação do reservatório. Espécies foram identificadas por inspeção manual e classificação semi-automática das gravações. Para acessar a similaridade entre a vegetação a jusante e montante, utilizamos composições de imagens Landsat TM/ETM+ de antes (2009-2011) e após (2016-2018) a formação do reservatório. Sítios de floresta de várzea foram similares antes, mas não após o reservatório. Sítios de vegetação sucessional inicial já diferiam antes do reservatório. Registramos 195 espécies de aves. A riqueza de espécies não diferiu entre os sítios a jusante e montante, mas a composição de espécies diferiu significativamente. Dez espécies foram indicadoras de vegetação sucessional inicial a montante e quatro a jusante. Dez espécies foram indicadoras de floresta de várzea a montante e 31 a jusante. Sete de 24 espécies especialistas de várzea foram detectadas apenas a montante pelas classificações semi-automáticas. Encontramos algumas espécies típicas de vegetação sucessional inicial a montante, porém não encontramos a maioria de espécies típicas the floresta alta de várzea. Predominantemente, aves carnívoras, insetívoras e nectarívoras aparentam ter sido substituídas por espécies generalistas e amplamente distribuídas.

Impacts of a large hydroelectric dam on the Madeira River (Brazil) on floodplain avifauna / Impacto de uma grande usina hidroelétrica sobre a avifauna de várzea doRio Madeira (Brasil)

Hydroelectric dams represent an important threat to seasonally flooded environments in the Amazon basin. We aimed to evaluate how a dam in the Madeira River, one of the largest tributaries of the Amazonas River, affected floodplain avifauna. Bird occurrence was recorded through simultaneous passive acoustic monitoring in early successional vegetation and floodplain forest downstream from the dam and upstream in sites impacted by permanent flooding after dam reservoir filling. Species were identified through manual inspection and semi-automated classification of the recordings. To assess the similarity in vegetation between downstream and upstream sites, we used Landsat TM/ETM+ composite images from before (2009-2011) and after (2016-2018) reservoir filling. Downstream and upstream floodplain forest sites were similar before, but not after dam construction. Early successional vegetation sites were already different before dam construction. We recorded 195 bird species. While species richness did not differ between upstream and downstream sites, species composition differed significantly. Ten species were indicators of early successional vegetation upstream, and four downstream. Ten species were indicators of floodplain forest upstream, and 31 downstream. Seven of 24 floodplain specialist species were detected by the semi-automated classification only upstream. While we found some bird species characteristic of early successional vegetation in the upstream sites, we did not find most species characteristic of tall floodplain forest. Predominantly carnivorous, insectivorous, and nectarivorous species appear to have been replaced by generalist and widely distributed species.(AU)

Barragens hidroelétricas representam uma importante ameaça a ambientes sazonalmente alagados na Amazônia. Avaliamos como uma barragem no Rio Madeira, um dos maiores tributários do Rio Amazonas, afetou a comunidade de aves de várzea. A ocorrência de aves foi registrada através de monitoramento acústico passivo simultâneo em vegetação em estágio sucessional inicial e floresta de várzea a jusante e em áreas a montante alagadas permanentemente após a formação do reservatório. Espécies foram identificadas por inspeção manual e classificação semi-automática das gravações. Para acessar a similaridade entre a vegetação a jusante e montante, utilizamos composições de imagens Landsat TM/ETM+ de antes (2009-2011) e após (2016-2018) a formação do reservatório. Sítios de floresta de várzea foram similares antes, mas não após o reservatório. Sítios de vegetação sucessional inicial já diferiam antes do reservatório. Registramos 195 espécies de aves. A riqueza de espécies não diferiu entre os sítios a jusante e montante, mas a composição de espécies diferiu significativamente. Dez espécies foram indicadoras de vegetação sucessional inicial a montante e quatro a jusante. Dez espécies foram indicadoras de floresta de várzea a montante e 31 a jusante. Sete de 24 espécies especialistas de várzea foram detectadas apenas a montante pelas classificações semi-automáticas. Encontramos algumas espécies típicas de vegetação sucessional inicial a montante, porém não encontramos a maioria de espécies típicas the floresta alta de várzea. Predominantemente, aves carnívoras, insetívoras e nectarívoras aparentam ter sido substituídas por espécies generalistas e amplamente distribuídas.(AU)

Dataset on audio records of animals from the northeast Andes of Colombia I: The bird sounds of Boyacá department.

This dataset is the first effort to combine the audio biodiversity of a taxonomic group in a selected location, the Boyacá department in Colombia. We conducted a detailed review of the sound recordings for birds from the Boyacá department within three repositories, the environmental sound collection of the Humboldt Institute, the Macaulay Library of the Cornell Lab of Ornithology, and the xeno-canto platform of the Naturalis Biodiversity Center. We selected recordings that were identified up to species and had complete metadata information. Using latitude and longitude information, we assigned each recording to one of the three regions and one of the 12 biotic units reported for Boyacá. We reported a total of 2321 recordings belonging to the Andean region (1892), Orinoquian region (425), and Carare-Lebrija-Nechi-Sinu (4). The sounds of Boyacá birds have been sampled for approximately three decades, with two peaks of activity in the early 2000's and 2018. We also included a map with the distribution of biotic units and sound recordings of our dataset. This dataset can be used to extract acoustic traits to test hypotheses of turnover in the acoustic space or traits by species, or to compare acoustic traits between species. It can also allow decision-makers to support biodiversity-based economies such as avitourism.

Explosive breeding in tropical anurans: environmental triggers, community composition and acoustic structure.

BACKGROUND: Anurans largely rely on acoustic communication for sexual selection and reproduction. While multiple studies have focused on the calling activity patterns of prolonged breeding assemblages, species that concentrate their reproduction in short-time windows, explosive breeders, are still largely unknown, probably because of their ephemeral nature. In tropical regions, multiple species of explosive breeders may simultaneously aggregate leading to massive, mixed and dynamic choruses. To understand the environmental triggers, the phenology and composition of these choruses, we collected acoustic and environmental data at five ponds in French Guiana during a rainy season, assessing acoustic communities before and during explosive breeding events. RESULTS: We detected in each pond two explosive breeding events, lasting between 24 and 70 h. The rainfall during the previous 48 h was the most important factor predicting the emergence of these events. During explosive breeding events, we identified a temporal factor that clearly distinguished pre- and mid-explosive communities. A common pool of explosive breeders co-occurred in most of the events, namely Chiasmocleis shudikarensis, Trachycephalus coriaceus and Ceratophrys cornuta. Nevertheless, the species composition was remarkably variable between ponds and for each pond between the first and the second events. The acoustic structure of explosive breeding communities had outlying levels of amplitude and unexpected low acoustic diversity, significantly lower than the communities preceding explosive breeding events. CONCLUSIONS: Explosive breeding communities were tightly linked with specific rainfall patterns. With climate change increasing rainfall variability in tropical regions, such communities may experience significant shifts in their timing, distribution and composition. In structurally similar habitats, located in the same region without obvious barriers, our results highlight the variation in composition across explosive breeding events. The characteristic acoustic structure of explosive breeding events stands out from the circadian acoustic environment being easily detected at long distance, probably reflecting behavioural singularities and conveying heterospecific information announcing the availability of short-lived breeding sites. Our data provides a baseline against which future changes, possibly linked to climate change, can be measured, contributing to a better understanding on the causes, patterns and consequences of these unique assemblages.

Multiscale spatio-temporal patterns of boat noise on U.S. Virgin Island coral reefs.

Sound-sensitive organisms are abundant on coral reefs. Accordingly, experiments suggest that boat noise could elicit adverse effects on coral reef organisms. Yet, there are few data quantifying boat noise prevalence on coral reefs. We use long-term passive acoustic recordings at nine coral reefs and one sandy comparison site in a marine protected area to quantify spatio-temporal variation in boat noise and its effect on the soundscape. Boat noise was most common at reefs with high coral cover and fish density, and temporal patterns reflected patterns of human activity. Boat noise significantly increased low-frequency sound levels at the monitored sites. With boat noise present, the peak frequencies of the natural soundscape shifted from higher frequencies to the lower frequencies frequently used in fish communication. Taken together, the spectral overlap between boat noise and fish communication and the elevated boat detections on reefs with biological densities raises concern for coral reef organisms.