Acoustic Identification Research Articles

Improving acoustic species identification using data augmentation within a deep learning framework

Convolutional neural networks (CNNs) are effective tools for acoustic classification tasks such as species identification. Large datasets of labelled recordings are required to develop CNN classifiers which can be difficult to obtain, particularly if species are rare or vocalise infrequently. Additionally, data often requires manual labelling which can be time consuming requiring expert analysis. Artificially generating data using augmentation can address these challenges, however the impact of data augmentation on CNN performance is poorly understood and often omitted in bioacoustic studies. Here, we empirically test the impact of CNN architecture and 20 data augmentation methods on classifier performance. We use acoustic identification of 18 small mammal species as a case study of a species group that can be effectively surveyed by acoustic monitoring, but recordings for training data are scarce and difficult to collect. Networks that achieved the highest accuracy across all sample sizes was a 10-layer CNN (96.43 %) and a pre-trained ResNet50 model (96.37 %). Overall, all augmentation effects improved ResNet50 model performance and 17 effects improved Conv10 performance, increasing relative change in accuracy (RCA) by 0.021–0.641. Three augmentation effects negatively impacted Conv10 RCA by −0.042 to −0.182. We also show that adding augmented data when the number of original samples is low has the greatest positive impact on accuracy and this effect was larger with ResNet50 models. Our work demonstrates that using data augmentation where few original samples are available can considerably improve model performance and highlights the potential of augmentation in developing acoustic classifiers for species where data are limited or difficult to obtain.

Hierarchical-taxonomy-aware and attentional convolutional neural networks for acoustic identification of bird species: A phylogenetic perspective

The study of bird populations is crucial for biodiversity research and conservation. Deep artificial neural networks have revolutionized bird acoustic recognition; however, most methods overlook inherent relationships among bird populations, resulting in the loss of biological information. To address this limitation, we propose the Phylogenetic Perspective Neural Network (PPNN), which incorporates hierarchical multilevel labels for each bird. PPNN uses a hierarchical semantic embedding framework to capture feature information at different levels. Attention mechanisms are employed to extract and select common and distinguishing features, thereby improving classification accuracy. We also propose a path correction strategy to rectify inconsistent predictions. Experimental results on bird acoustic datasets demonstrate that PPNN outperforms current methods, achieving classification accuracies of 90.450%, 91.883%, and 89.950% on the Lishui-Zhejiang Birdsdata (100 species), BirdCLEF2018-Small (150 species), and BirdCLEF2018-Large (500 species) datasets respectively, with the lowest hierarchical distance of a mistake across all datasets. Our proposed method is applicable to any bird acoustic dataset and presents significant advantages as the number of categories increases.

Ultrasound-Powered Wireless Underwater Acoustic Identification Tags for Backscatter Communication.

Autonomous underwater vehicle (AUV) operations are limited by currently achievable underwater localization and navigation solutions; hence, the development of low-cost and passive (i.e., operable without an active power supply) acoustic underwater markers (or tags) can provide accurate localization information to AUVs improving their situational awareness, especially when operating in small scales or confined missions. This work presents an acoustic identification (AID) tag that can be powered wirelessly with ultrasonic power transfer from a remote acoustic source (e.g., mounted on an interrogating AUV) and provide localization information using backscatter communication. The AID tag harvests energy from the acoustic signal generated from the AUV and communicates by modulating the reflected signals from an embedded piezoelectric transducer. A scaled broadband AID tag prototype that achieves concurrent acoustic energy harvesting (tuned around 1.3 MHz) and backscatter communication (in wider frequency band 600 and 800 kHz) using frequency-domain multiplexing is implemented using a custom broadband impedance matching-based transducer design approach. During concurrent power and data operation, this prototype AID tag achieves data rates up to 200 kb/s using amplitude- and frequency-based modulation communication. The use of broadband schemes to achieve robust communications in low SNR (tested here down to -6 dB) is also demonstrated using linear frequency-modulated data carriers. Finally, the extension to full-scale devices of this AID tag concept and potential applications for short-range AUV routing and navigation such as homing and docking are discussed.

Remarks on the diversity and echolocation calls of hipposiderid bats (Chirroptera: Hipposideridae) in Cuc Phuong National Park, northeastern Vietnam

Cuc Phuong is the oldest and most well-known National Park in Vietnam. With its large primary forest area and karst cave systems, it is assumed to be an ideal home to a highly diverse bat fauna. However, species diversity and echolocation of bats in this park are poorly characterized. Between April 2021 and April 2022, three bat surveys were conducted in the park with special attention on the species of the family Hipposideridae. The bat specimens collected previously from the park were also examined. According to past records, hipposiderids in the park comprised at least seven species: Aselliscus stoliczkanus, Coelops frithii, Hipposideros alongensis, Hipposideros armiger, Hipposideros gentilis, Hipposideros lylei and Hipposideros poutensis. In this series of field surveys, six of these seven species were captured while Coelops frithii was only confirmed from specimen examination. Each species is clearly distinguishable by their morphology and echolocation calls. This study provides the first comparison of echolocation calls of all seven species which will be useful for potential acoustic identification and monitoring bats in the park and surrounding areas in future.

AFPILD: Acoustic footstep dataset collected using one microphone array and LiDAR sensor for person identification and localization

Empowering the auditory perception-based person identification and localization could enable the service robots to navigate among the crowds in a more socially compliant manner. And it is more user-friendly with fewer privacy concerns compared to the widely studied camera-supported approaches. One way to achieve such auditory person identification and localization is to identify and locate the subject by hearing her/his footstep. As far as we know, although several related acoustic identification benchmark datasets have been made, the dataset that simultaneously concentrates on acoustic identification and localization is absent. Aiming at the shortage of such benchmark datasets, we propose to build the acoustic footstep-based person identification and localization dataset (AFPILD) by unifying the identify and locate tasks for the first time, concerning the clothing and shoe type covariates from the human-aware mobile service robot navigation perspective. Specifically, we record the walking sounds of 40 subjects with a lightweight microphone array in one real unstrained indoor room. For ground truth localization label annotation, we record the point clouds of the walking subjects with a multi-line LiDAR synchronously. The AFPILD contains nearly 10 h of footstep audios (around 62,174 footstep events). To the best of our knowledge, this is the largest and most comprehensive conducted acoustic footstep dataset built for person identification and localization study up to now. We also present the baseline works concerning the identification, localization, and synchronous identify-locate tasks, respectively, to compare various potential approaches on this dataset in the future. Experimental results show that the covariates of clothing and shoe type degrade the footstep identification (around 12% of clothing and 33% of shoe type), while their impacts on the localization are unobvious.

CNN-based acoustic identification of gas–liquid jet: Evaluation of noise resistance and visual explanation using Grad-CAM

For the analysis of anomalies in a steam generator (SG) of a sodium-cooled fast reactor (SFR), we evaluate the noise resistance of CNN-based acoustic identification methods of gas–liquid two-phase jets and produce visual explanations for their decisions. First, we introduce the water flow sound and the three types of gas–liquid jet sounds, which simulate the background noise and the anomaly sounds, respectively. Second, we produce time–frequency representations for various signal-to-noise ratios (SNRs) and employ AlexNet, VGG16, and ResNet18 to the identification of the gas–liquid two-phase jets. As a result, the best CNN of ResNet18 achieves more than 92% for SNR=0,−4,−8,and−12 dB and 69% for SNR=−16and−20 dB. This result indicates that our proposed methods can identify the flow states of gas–liquid two-phase jets in low-level noise environments and detect the gas–liquid two-phase jets even in high-level noise environments. Also, Grad-CAM suggests that ResNet18 focuses on one of the spectrum peaks of the water flow sound and all or part of the signal intensity pattern of the gas–liquid jet sounds. Our proposed methods lead to the safe operation and fast, accurate, and accountable analysis of anomalies in SFR.

Bats of the Tunisian Desert: Preliminary Data Using Acoustic Identification and First Record of Taphozous nudiventris in the Country

Increasing aridity usually results in decreasing bat abundance and species richness, and the Saharan desert is an example of such impoverishment. Moreover, the harsh climatic and field conditions in this area restrict the feasibility of surveys. Therefore, the bat fauna of the Tunisian Sahara was unstudied until an expedition was conducted in May 2021. A total of seven species were recorded using passive bat detectors set mainly at water bodies, which concentrate bat activity. Echolocation calls of these species did not depart from published records for the Mediterranean area. Our data failed to identify the two ecomorphotypes of Pipistrellus kuhlii but confirmed the adaptability of this species to local conditions. For the first time in Tunisia, we recorded echolocation calls of Taphozous nudiventris, a species that has been rarely reported in North Africa. The highest species richness was identified at the largest wetlands of Oued Daghsen (Oued Dkouk Nature Reserve) and at Bordj el Khadra (only six and four species, respectively). Most Mediterranean species were recorded in the northern locality, while desert species were more active in the most southern one. The importance of water bodies for bats was confirmed once more. No sign of competition among species was detected, as the Mediterranean and desert-dwelling bat species were active at the same time. Additional surveys should be planned in order to enlarge the list of seven species recorded in only one week.

Modelling andin situobservation of broadband acoustic scattering from the Silver cyprinid (Rastrineobola argentea)in Lake Victoria, East Africa

AbstractLake Victoria is the second-largest freshwater lake in the world, and fish from the lake are a vital food resource for millions of people living around it. The silver cyprinid (Rastrineobola argentea), a small schooling pelagic species known in Tanzania as “dagaa” contributes ca. 55% to the total annual catch (ca. 0.51 million tonnes (MT) in 2014). The acoustic target strength (TS, dB re 1 m2) of dagaa, a key factor for biomass estimation, is however not well described, and is a major source of uncertainty in biomass estimation. In this study, we developed a Kirchhoff-ray mode (KRM) model to predict the TS of dagaa at standard fisheries survey frequencies. The model was based on the morphology of the body and the dual-chambered swimbladder, as obtained from X-ray images of fish ranging in total length (TL) between 2.8 and 5.4 cm. The results suggested that the swimbladder (which comprises 2.6 to 8.2% of body volume) accounts for ca. 65 to 90% of the total backscattering at 120 kHz. The predicted TS was highly dependent on tilt angle, varying by 14.0 dB at 120 kHz across the tilt range 65–115˚ (where 0˚ is head up and 180˚ is tail up), and TS variability with tilt generally increased with increasing frequency. The tilt angle of acoustically tracked individual fish indicated a distribution of tilt angles with a mean and s.d. of 93.5 and 15.1°. Our model suggested a new tilt-averaged TS–TL relationship for dagaa [$T{S}_{120\ kHz} = 19.49\log ( {TL} ) - 70.3$], which leads to a TS 1.5 dB higher than the value in the relationship presently used to estimate stock biomass. The new relationship will lead to a substantial reduction (by ca. 30%) in estimated biomass. The discrepancies between the mean relative frequency response of the in situ measurements of backscatter from dagaa and the KRM model predictions were in the range of -2.9–3.1 dB at frequencies from 45 to 250 kHz. The KRM modelling and in situ broadband measurements of dagaa will be beneficial for acoustic identification and behavioural studies of dagaa, and will enable improved biomass assessment, thereby underpinning sustainable long-term management.

Acoustic Identification of Dolphin Whistle Types in Deep Waters of Arabian Sea Using Wavelet Threshold Denoising Approach

Acoustic Identification of Dolphin Whistle Types in Deep Waters of Arabian Sea Using Wavelet Threshold Denoising Approach

Nondestructive characterization and artificial intelligence recognition of acoustic identifiers of ancient ceramics

Cultural heritage identity management is the most basic and important work in the process of cultural heritage protection. It is of great significance to provide a unique and identifiable digital identity for ancient ceramics. At present, the identification information of ancient ceramics is mainly composed of external visual characteristics, and there is no report on feature identification method that can reflect the properties of ancient ceramics. Audible sound signals not only have advantages in non-destructive testing, but also can be used as voiceprint information to identify, monitor and analyze ancient ceramics. In this paper, seven ancient ceramics and 12 similar modern ceramic cups are taken as research objects, and an acoustic identifier (AID) is constructed. We put forward a reliable acoustic identification method for ancient ceramics, and established a digital code of acoustic characteristics of ancient ceramics. The results show that audible sound waves can reflect the attribute information of ancient ceramics. Sufficient acoustic data combined with deep learning can not only accurately match the identity of ancient ceramics, but also detect the real-time identity information of ancient ceramics, and make a comparative analysis of its cracks and whether it has caused damage. This method can provide a variety of practical applications for audible signal feature recognition technology in the exhibition, protection, trading, recognition and safety management of ancient ceramics and other cultural relics.

Characterization of unmanned aircraft system acoustic noise

UAS (unmanned aircraft systems) are used in a variety of commercial and military enterprises. In military settings, being able to identify location as well as number of enemy UAS can lead to stronger strategic positioning or appropriate countermeasures. In community settings, UAS can be a source of annoyance. Recently, AFRL measured three different UAS flown by pilots from Sinclair Community College. These UAS included various sizes and blade amounts. They were flown back and forth through an array of 20 acoustic sensors at nominal thrust. In addition to flying through the array, they were also flown at heights of 15 feet and 100 ft above the array. Using both telemetry and acoustic data, spherical source characterizations of the systems can be derived. These three-dimensional characterizations can lead to a better understanding of UAS acoustics as well as improved measurement design. This study describes the recent measurement in detail, some preliminary analysis, and the lessons learned. Future analysis and testing will promote acoustic identification of a larger variety of UAS.UAS (unmanned aircraft systems) are used in a variety of commercial and military enterprises. In military settings, being able to identify location as well as number of enemy UAS can lead to stronger strategic positioning or appropriate countermeasures. In community settings, UAS

High resolution acoustic identification of clusters of small blockages in fluid-filled pipe using maximum likelihood estimation

This paper presents a method for identifying a cluster of small blockages (i.e., blockages with length on the order of centimeters and radial extent on the order of millimeters and separated by a distance on the order of few centimeters) in pressurized fluid-filled pipes using sound waves. This focus on defects with small scale, and, thus, small scattering strength is exploited to develop a Neumann series solution for the scattered acoustic wave field. The probing waves are such that the Helmholtz number (ratio of blockage longitudinal length scale and probing acoustic wavelength) is of order 1 or larger. A high resolution inverse technique for identifying a cluster of small blockages based on the maximum likelihood estimation principle is developed. The proposed technique uses two-dimensional search space to resolve each blockage in the cluster and requires a single measurement point only. The method is successfully tested through both numerical and laboratory experiments. The proposed methodology allows an early identification of a cluster of small defects and leads to reliable condition assessment of pipelines, which is necessary to inform decisions as to when remedial actions are required.

Compact and automated eDNA sampler for in situ monitoring of marine environments

Using environmental DNA (eDNA) to monitor biodiversity in aquatic environments is becoming an efficient and cost-effective alternative to other methods such as visual and acoustic identification. Until recently, eDNA sampling was accomplished primarily through manual sampling methods; however, with technological advances, automated samplers are being developed to make sampling easier and more accessible. This paper describes a new eDNA sampler capable of self-cleaning and multi-sample capture and preservation, all within a single unit capable of being deployed by a single person. The first in-field test of this sampler took place in the Bedford Basin, Nova Scotia, Canada alongside parallel samples taken using the typical Niskin bottle collection and post-collection filtration method. Both methods were able to capture the same aquatic microbial community and counts of representative DNA sequences were well correlated between methods with R^{2} values ranging from 0.71–0.93. The two collection methods returned the same top 10 families in near identical relative abundance, demonstrating that the sampler was able to capture the same community composition of common microbes as the Niskin. The presented eDNA sampler provides a robust alternative to manual sampling methods, is amenable to autonomous vehicle payload constraints, and will facilitate persistent monitoring of remote and inaccessible sites.

Density-depended Acoustical Identification of Two Common Seaweeds (Posidonia Oceanica and Cymodocea Nodosa) in the Mediterranean Sea

The non-destructive samplings are very important in not damaging seagrasses and seaweed under protection, at the field studies. The grasses are prominent in the assessment of the ecological status of the marine environments. One of the effective non-destructive samplings was the acoustical methods which need a low level of the sea and atmospheric conditions as compared to the other remote sensing system. Like the others, acoustic data alone are inherently ambiguous concerning the identities of the scatterers and need sea-truthing at the field studies. Considering the requirements above and the advantages of the acoustical methods, an acoustical in situ study was conducted to discriminate two dominant seagrasses along the western Turkish Mediterranean coast in August (in the year 2012) when biometrics of both species was at maxima. Four different regions were involved in the study during data collection using a split beam echosounder operated at 206 kHz since each region had different strengths of their biometrics reflecting regional variations of the acoustical measurements. For discrimination, a statistical approach based on the acoustical parameters and their statistics was enriched and validated with multivariate analyses (Silhouette, k-means, PCA and CAP analyses). Posidonia oceanica was correlated with mean, median, SD and maximum value of Sa, whereas Cymodocea nodosa was characterized by hardness and roughness of leaf echo, followed by skewness and kurtosis of Sa. The acoustics of C. nodosa were related to regional differences, but P. oceanica was related to the depthwise difference. Both species had highly different densities (g/cm3), followed by biomass. The present study will interactively help acoustical studies to be more common and comprehensible and the other previous studies published did as well.

Passive underwater Acoustic IDentification (AID) tags for enhancing Autonomous Underwater Vehicle (AUV) navigation during docking or homing operations

Autonomous underwater vehicle (AUV) navigation requires accurate positioning information from the surrounding environment, especially for tasks such as AUV homing or docking operations. Previous literature has introduced a class of low-cost passive underwater acoustic markers, termed Acoustic IDentification (AID) tags [Satish and Sabra, J. Acoust. Soc. Am. 149(5), 3387–3405 (2021)] built of multi-layer shells with different acoustic properties and thicknesses to generate a uniquely engineered acoustic signature, composed of the multiple reflections created by the layer interfaces. These passive AID tags can be detected by an AUV instrumented with a high-frequency sonar transducer at significantly greater distances than conventional optical methods, especially in turbid waters. Additionally using AID tags as navigation-aid can also alleviate the need of relying on active acoustic transponders. An implementation of a constellation of AID tags enabling fine underwater positioning an AUV towards a docking station or for homing purposes will be presented to provide proof of concept. Furthermore, the optimization of the design of the AID tags for this application aswell as specific signal processing detection methodologies to improve thedetectability of AID tags in the presence of interfering signals (e.g., clutter) will be discussed. [Work supported by the Office of Naval Research].

Improving passive acoustic target detection using machine learning classifiers

The ocean covers 70% of the Earth surface area, yet over 80% of it remains unexplored despite the advances in underwater acoustics and oceanography. Ocean exploration is critical for accurate climate model development, renewable energy applications, and in understanding the marine habitat. Further exploration necessitates improvements in underwater navigation with Autonomous Underwater Vehicles (AUVs). Utilizing acoustic landmarks can enhance AUV localization performance [Fula et al., Oceans IEEE/MTS (2018)]. Passive markers called Acoustic Identification (AID) Tags have unique and identifiable acoustic reflection signatures designed to function as landmarks [Satish et al., JASA 149 (2021)]. These targets can be detected by Match Filtering returns with template signals. Match Filtering performance is limited in the presence of strong interfering signals, and with changes in sound speeds that alters the temporal structure of these signatures. The application of a Machine Learning classifiers for detecting AID tag signatures can improve the localization performance. Through implementation of Logistic Regression, Deep Neural Networks, and Convolutional Neural Networks, the generalizability and superiority of these models is demonstrated. We report high accuracies (>90 %) in a multilabel (8 classes) classification task with signals with low SNR (–6 dB) and strong interference (+12 dB).

Optimizing wireless acoustic energy harvesting and communication with passive markers underwater

To enhance functionality of Acoustic Identification Tags (AID), passive markers designed as landmarks for Autonomous Underwater Vehicles (AUVs) [Satish et al., JASA 149 (2021)], the development of an accompanying battery free active communication platform was explored [Bhardwaj et al., JASA 152, 4 (2022)]. This platform utilizes custom piezoelectric transducers impedance matched for broadband operation, enabling concurrent acoustic energy harvesting and fast backscatter communication [Allam et al., SPIE 31, 9 (2022)]. The combination of these active and passive devices can improve AUV localization and information transfer for specific applications such as short-range search missions or AUV homing and docking. The design of this platform for underwater operations requires a balance of parameters such as center frequency and piezoelectric element dimensions to achieve a high range of operation, power efficiency and data rates, while minimizing attenuation and spreading losses challenges. The design process for piezo selection, acoustic and electrical impedance matching, transducer casing design, and the experimental evaluation of the chosen 350 kHz design will be discussed. This approach was validated experimentally to quantify performance metrics such as the maximal range, data rates using various standard communication such as Frequency Shift Keying (FSK) and Amplitude Modulation (AM).

A review of coal gangue identification research-application to China's top coal release process.

Top coal caving is a process for the rational extraction of large amounts of coal resources. However, this process readily causes release of excessive amounts of gangue during the coal release process. The conventional technique, which involves visual inspection, is not only labor-intensive but also can introduce inaccuracies. Coal and gangue identification methods used during the top coal caving process are reviewed herein. The methods developed and studied over the last 20years, such as gamma identification, acoustic identification, radar detection, vibration technology, image identification, and infrared positioning, are described separately, and their principles for use are explained. The advantages and disadvantages of these methods are analyzed, and an evaluation index for coal gangue identification methods is established by using hierarchical analysis (AHP). The results indicate that the γ-ray method is optimal to use under the current system. A future gangue identification method is conceived and proposed, which has reference value for the research and applications of gangue identification technology.

Acoustic identification and classification of four dolphin species in the Brazilian marine area affected by the largest tailings dam failure disaster.

Passive acoustic monitoring (PAM) is an increasingly used technique to access the occurrence, distribution, and abundance of cetaceans that may be visually unavailable most of the time. The largest tailings dam failure disaster occurred on 5 November 2015, when the Fundão dam collapsed, releasing over 50 million cubic meters of tailings into the Doce River basin; 14 days later, the tailings plume reached the Atlantic Ocean. PAM was implemented in the concerned area and cetacean species were acoustically identified. Whistles and clicks of visual and acoustic matches were used to predict and classify exclusive acoustic records through random forest models. The identified species were Guiana, rough-toothed, and bottlenose dolphins. Additionally, the franciscana, the most threatened cetacean in the western South Atlantic Ocean, was also acoustically identified. The whistle classifier had 86.9% accuracy with final frequency, duration, and maximum frequency ranked as the most important parameters. The clicks classifier had 86.7% accuracy with peak frequency and 3 dB bandwidth as the most important parameters for classifying species. Considering the potential effect of the increase in turbidity on sound transmission, such as attenuation, the presented classifier should be continuously improved with novel data collected from long-term acoustic monitoring.

Acoustic event characterization for service robot using convolutional networks

<p>This paper presents and discusses the creation of a sound event classification model using deep learning. In the design of service robots, it is necessary to include routines that improve the response of both the robot and the human being throughout the interaction. These types of tasks are critical when the robot is taking care of children, the elderly, or people in vulnerable situations. Certain dangerous situations are difficult to identify and assess by an autonomous system, and yet, the life of the users may depend on these robots. Acoustic signals correspond to events that can be detected at a great distance, are usually present in risky situations, and can be continuously sensed without incurring privacy risks. For the creation of the model, a customized database is structured with seven categories that allow to categorize a problem, and eventually allow the robot to provide the necessary help. These audio signals are processed to produce graphical representations consistent with human acoustic identification. These images are then used to train three convolutional models identified as high-performing in this type of problem. The three models are evaluated with specific metrics to identify the best-performing model. Finally, the results of this evaluation are discussed and analyzed.</p>