Ecological inferences about marine mammals from passive acoustic data.

Monitoring on the basis of sound recordings, or passive acoustic monitoring, can complement or serve as an alternative to real-time visual or aural monitoring of marine mammals and other animals by human observers. Passive acoustic data can support the estimation of common, individual-level ecological metrics, such as presence, detection-weighted occupancy, abundance and density, population viability and structure, and behaviour. Passive acoustic data also can support estimation of some community-level metrics, such as species richness and composition. The feasibility of estimation and certainty of estimates is highly context dependent, and understanding the factors that affect the reliability of measurements is useful for those considering whether to use passive acoustic data. Here, we review basic concepts and methods of passive acoustic sampling in marine systems that often are applicable to marine mammal research and conservation. Our ultimate aim is to facilitate collaboration among ecologists, bioacousticians, and data analysts. Ecological applications of passive acoustics require one to make decisions about sampling design, which in turn requires consideration of sound propagation, sampling of signals, and data storage. One also must make decisions about signal detection and classification and evaluation of the performance of algorithms for these tasks. Investment in the research and development of systems that automate detection and classification, including machine learning, are increasing. Passive acoustic monitoring is more reliable for detection of species presence than for estimation of other species-level metrics. Use of passive acoustic monitoring to distinguish among individual animals remains difficult. However, information about detection probability, vocalisation or cue rate, and relations between vocalisations and the number and behaviour of animals increases the feasibility of estimating abundance or density. Most sensor deployments are fixed in space or are sporadic, making temporal turnover in species composition more tractable to estimate than spatial turnover. Collaborations between acousticians and ecologists are most likely to be successful and rewarding when all partners critically examine and share a fundamental understanding of the target variables, sampling process, and analytical methods.

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.

Tracking time differences of arrivals of multiple sound sources in the presence of clutter and missed detections.

Acoustic line transect surveys are often used in combination with visual methods to estimate the abundance of marine mammal populations. These surveys typically use towed linear hydrophone arrays and estimate the time differences of arrival (TDOAs) of the signal of interest between the pairs of hydrophones. The signal source TDOAs or bearings are then tracked through time to estimate the animal position, often manually. The process of estimating TDOAs from data and tracking them through time can be especially challenging in the presence of multiple acoustically active sources, missed detections, and clutter (false TDOAs). This study proposes a multi-target tracking method to automate TDOA tracking. The problem formulation is based on the Gaussian mixture probability hypothesis density filter and includes multiple sources, source appearance and disappearance, missed detections, and false alarms. It is shown that by using an extended measurement model and combining measurements from broadband echolocation clicks and narrowband whistles, more information can be extracted from the acoustic encounters. The method is demonstrated on false killer whale (Pseudorca crassidens) recordings from Hawaiian waters.

Model-based localization of deep-diving cetaceans using towed line array acoustic data.

Passive acoustic monitoring using a towed line array of hydrophones is a standard method for localizing cetaceans during line-transect cetacean abundance surveys. Perpendicular distances estimated between localized whales and the trackline are essential for abundance estimation using acoustic data. Uncertainties in the acoustic data from hydrophone movement, sound propagation effects, errors in the time of arrival differences, and whale depth are not accounted for by most two-dimensional localization methods. Consequently, location and distance estimates for deep-diving cetaceans may be biased, creating uncertainty in abundance estimates. Here, a model-based localization approach is applied to towed line array acoustic data that incorporates sound propagation effects, accounts for sources of error, and localizes in three dimensions. The whale's true distance, ship trajectory, and whale movement greatly affected localization results in simulations. The localization method was applied to real acoustic data from two separate sperm whales, resulting in three-dimensional distance and depth estimates with position bounds for each whale. By incorporating sources of error, this three-dimensional model-based approach provides a method to address and integrate the inherent uncertainties in towed array acoustic data for more robust localization.

Improve automatic detection of animal call sequences with temporal context.

Many animals rely on long-form communication, in the form of songs, for vital functions such as mate attraction and territorial defence. We explored the prospect of improving automatic recognition performance by using the temporal context inherent in song. The ability to accurately detect sequences of calls has implications for conservation and biological studies. We show that the performance of a convolutional neural network (CNN), designed to detect song notes (calls) in short-duration audio segments, can be improved by combining it with a recurrent network designed to process sequences of learned representations from the CNN on a longer time scale. The combined system of independently trained CNN and long short-term memory (LSTM) network models exploits the temporal patterns between song notes. We demonstrate the technique using recordings of fin whale (Balaenoptera physalus) songs, which comprise patterned sequences of characteristic notes. We evaluated several variants of the CNN + LSTM network. Relative to the baseline CNN model, the CNN + LSTM models reduced performance variance, offering a 9-17% increase in area under the precision-recall curve and a 9-18% increase in peak F1-scores. These results show that the inclusion of temporal information may offer a valuable pathway for improving the automatic recognition and transcription of wildlife recordings.

Publisher Correction: Deep neural networks for automated detection of marine mammal species.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Deep neural networks for automated detection of marine mammal species.

Deep neural networks have advanced the field of detection and classification and allowed for effective identification of signals in challenging data sets. Numerous time-critical conservation needs may benefit from these methods. We developed and empirically studied a variety of deep neural networks to detect the vocalizations of endangered North Atlantic right whales (Eubalaena glacialis). We compared the performance of these deep architectures to that of traditional detection algorithms for the primary vocalization produced by this species, the upcall. We show that deep-learning architectures are capable of producing false-positive rates that are orders of magnitude lower than alternative algorithms while substantially increasing the ability to detect calls. We demonstrate that a deep neural network trained with recordings from a single geographic region recorded over a span of days is capable of generalizing well to data from multiple years and across the species' range, and that the low false positives make the output of the algorithm amenable to quality control for verification. The deep neural networks we developed are relatively easy to implement with existing software, and may provide new insights applicable to the conservation of endangered species.

Depth- and range-dependent variation in the performance of aquatic telemetry systems: understanding and predicting the susceptibility of acoustic tag-receiver pairs to close proximity detection interference.

BACKGROUND: Passive acoustic telemetry using coded transmitter tags and stationary receivers is a popular method for tracking movements of aquatic animals. Understanding the performance of these systems is important in array design and in analysis. Close proximity detection interference (CPDI) is a condition where receivers fail to reliably detect tag transmissions. CPDI generally occurs when the tag and receiver are near one another in acoustically reverberant settings. Here we confirm transmission multipaths reflected off the environment arriving at a receiver with sufficient delay relative to the direct signal cause CPDI. We propose a ray-propagation based model to estimate the arrival of energy via multipaths to predict CPDI occurrence, and we show how deeper deployments are particularly susceptible. METHODS: A series of experiments were designed to develop and validate our model. Deep (300 m) and shallow (25 m) ranging experiments were conducted using Vemco V13 acoustic tags and VR2-W receivers. Probabilistic modeling of hourly detections was used to estimate the average distance a tag could be detected. A mechanistic model for predicting the arrival time of multipaths was developed using parameters from these experiments to calculate the direct and multipath path lengths. This model was retroactively applied to the previous ranging experiments to validate CPDI observations. Two additional experiments were designed to validate predictions of CPDI with respect to combinations of deployment depth and distance. Playback of recorded tags in a tank environment was used to confirm multipaths arriving after the receiver's blanking interval cause CPDI effects. RESULTS: Analysis of empirical data estimated the average maximum detection radius (AMDR), the farthest distance at which 95% of tag transmissions went undetected by receivers, was between 840 and 846 m for the deep ranging experiment across all factor permutations. From these results, CPDI was estimated within a 276.5 m radius of the receiver. These empirical estimations were consistent with mechanistic model predictions. CPDI affected detection at distances closer than 259-326 m from receivers. AMDR determined from the shallow ranging experiment was between 278 and 290 m with CPDI neither predicted nor observed. Results of validation experiments were consistent with mechanistic model predictions. Finally, we were able to predict detection/nondetection with 95.7% accuracy using the mechanistic model's criterion when simulating transmissions with and without multipaths. DISCUSSION: Close proximity detection interference results from combinations of depth and distance that produce reflected signals arriving after a receiver's blanking interval has ended. Deployment scenarios resulting in CPDI can be predicted with the proposed mechanistic model. For deeper deployments, sea-surface reflections can produce CPDI conditions, resulting in transmission rejection, regardless of the reflective properties of the seafloor.

Estimating and removing colorations from the deconvolved impulse response of an underwater acoustic channel.

The impulse response (IR) of an acoustic channel can be obtained by cross-correlating the received signal with the broadband excitation signal in unfavorable noise conditions. However, the deconvolved IR is colored by the IRs of the combined electrical equipment. This letter presents a time domain approach using pre-computed filters to whiten the unknown coloration in order to obtain the channel's time domain waveform. The method is validated with an image-source model and the IR of the channel is recovered with spectral root mean square error of -27 dB. Data results obtained from a pool experiment with non-calibrated equipment yield a whitened IR with standard deviation of 0.9 dB (30-68 kHz band).

Partial dereverberation used to characterize open circuit scuba diver signatures.

The use of passive acoustics to detect self-contained underwater breathing apparatus (SCUBA) divers is useful for nearshore and port security applications. While the performance of a detector can be optimized by understanding the signal's spectral characteristics, anechoic recording environments are generally not available or are cost prohibitive. A practical solution is to obtain the source spectra by equalizing the recording with the inverse of the channel's impulse response. This paper presents a dereverberation method for signal characterization that is subsequently applied to four recorded SCUBA configurations. The inverse impulse response is computed in the least-square sense, and partial dereverberation of SCUBA is performed over the 6-18 kHz band. Results indicate that early reflections and late reverberation added as much as 6.8 dB of energy. Mean unadjusted sound pressure levels computed over the 0.3-80 kHz band were 130 ± 5.9 dB re 1 µPa at 1 m. Bubble noise carries a significant amount of the total energy and masks the regulator signatures from 1.3 to 6 kHz, depending on the regulator configuration. While the dereverberation method is applied here to SCUBA signals, it is generally applicable to other sources if the impulse response of the recording environment can be obtained separately.

Estimating uncertainty in subsurface glider position using transmissions from fixed acoustic tomography sources.

Four acoustic Seagliders were deployed in the Philippine Sea November 2010 to April 2011 in the vicinity of an acoustic tomography array. The gliders recorded over 2000 broadband transmissions at ranges up to 700 km from moored acoustic sources as they transited between mooring sites. The precision of glider positioning at the time of acoustic reception is important to resolve the fundamental ambiguity between position and sound speed. The Seagliders utilized GPS at the surface and a kinematic model below for positioning. The gliders were typically underwater for about 6.4 h, diving to depths of 1000 m and traveling on average 3.6 km during a dive. Measured acoustic arrival peaks were unambiguously associated with predicted ray arrivals. Statistics of travel-time offsets between received arrivals and acoustic predictions were used to estimate range uncertainty. Range (travel time) uncertainty between the source and the glider position from the kinematic model is estimated to be 639 m (426 ms) rms. Least-squares solutions for glider position estimated from acoustically derived ranges from 5 sources differed by 914 m rms from modeled positions, with estimated uncertainty of 106 m rms in horizontal position. Error analysis included 70 ms rms of uncertainty due to oceanic sound-speed variability.

Methods for tracking multiple marine mammals with wide-baseline passive acoustic arrays.

Most methods used to track marine mammals with passive acoustics require that time differences of arrivals (TDOAs) are established and are associated between hydrophone pairs. Consequently, multiple animal trackers commonly apply single-animal TDOA localization methods after performing a call separation and/or TDOA association step. When a wide-baseline array is used with multiple animals that make similar calls with short inter-call-intervals, the separation/association step can be challenging and potentially rejects valid TDOAs. This paper extends a model-based TDOA method to deal with multiple-animal datasets in a way that does not require a TDOA association step; animals are separated based on position. Advantageously, false TDOAs (e.g., a direct path associated with a multipath arrival) do not need to be removed. An analogous development is also presented for a model-based time of arrival tracking method. Results from simulations and application to a multiple sperm whale dataset are used to illustrate the multiple-animal methods. Although computationally more demanding than most track-after-association methods because separation is performed in a higher-dimensional space, the methods are computationally tractable and represent a useful new tool in the suite of options available for tracking multiple animals with passive acoustics.

Sperm whale three-dimensional track, swim orientation, beam pattern, and click levels observed on bottom-mounted hydrophones.

In an earlier paper [Nosal and Frazer Appl. Acoust. 61, 1187-1201 (2006)], a sperm whale was tracked in three-dimensions using direct and surface-reflected time differences (DRTD) of clicks recorded on five bottom-mounted hydrophones, a passive method that is robust to timing errors between hydrophones. This paper refines the DRTD method and combines it with a time of (direct) arrival method to improve the accuracy of the track. The position and origin time of each click having been estimated, pitch and yaw are then obtained by assuming the main axis of the whale is tangent to the track. Roll is then found by applying the bent horn model of sperm whale phonation, in which each click is composed of two pulses, p0 and p1, that exit the whale at different points. With instantaneous pitch, roll, and yaw estimated from time differences, amplitudes are then used to estimate the beam patterns of the p0 and p1 pulses. The resulting beam patterns independently confirm those obtained by Zimmer et al. [J. Acoust. Soc. Am. 117, 1473-1485 (2005); 118, 3337-3345 (2005)] with a very different experimental setup. A method for estimating relative click levels is presented and used to find that click levels decrease toward the end of a click series, prior to the "creak" associated with prey capture.

Experimental evaluation of radiosity for room sound-field prediction.

An acoustical radiosity model was evaluated for how it performs in predicting real room sound fields. This was done by comparing radiosity predictions with experimental results for three existing rooms--a squash court, a classroom, and an office. Radiosity predictions were also compared with those by ray tracing--a "reference" prediction model--for both specular and diffuse surface reflection. Comparisons were made for detailed and discretized echograms, sound-decay curves, sound-propagation curves, and the variations with frequency of four room-acoustical parameters--EDT, RT, D50, and C80. In general, radiosity and diffuse ray tracing gave very similar predictions. Predictions by specular ray tracing were often very different. Radiosity agreed well with experiment in some cases, less well in others. Definitive conclusions regarding the accuracy with which the rooms were modeled, or the accuracy of the radiosity approach, were difficult to draw. The results suggest that radiosity predicts room sound fields with some accuracy, at least as well as diffuse ray tracing and, in general, better than specular ray tracing. The predictions of detailed echograms are less accurate, those of derived room-acoustical parameters more accurate. The results underline the need to develop experimental methods for accurately characterizing the absorptive and reflective characteristics of room surfaces, possible including phase.

Improved algorithms and methods for room sound-field prediction by acoustical radiosity in arbitrary polyhedral rooms.

This paper explores acoustical (or time-dependent) radiosity--a geometrical-acoustics sound-field prediction method that assumes diffuse surface reflection. The literature of acoustical radiosity is briefly reviewed and the advantages and disadvantages of the method are discussed. A discrete form of the integral equation that results from meshing the enclosure boundaries into patches is presented and used in a discrete-time algorithm. Furthermore, an averaging technique is used to reduce computational requirements. To generalize to nonrectangular rooms, a spherical-triangle method is proposed as a means of evaluating the integrals over solid angles that appear in the discrete form of the integral equation. The evaluation of form factors, which also appear in the numerical solution, is discussed for rectangular and nonrectangular rooms. This algorithm and associated methods are validated by comparison of the steady-state predictions for a spherical enclosure to analytical solutions.

Investigation of the validity of radiosity for sound-field prediction in cubic rooms.

This paper explores acoustical (or time-dependent) radiosity using predictions made in four cubic enclosures. The methods and algorithms used are those presented in a previous paper by the same authors [Nosal, Hodgson, and Ashdown, J. Acoust. Soc. Am. 116(2), 970-980 (2004)]. First, the algorithm, methods, and conditions for convergence are investigated by comparison of numerous predictions for the four cubic enclosures. Here, variables and parameters used in the predictions are varied to explore the effect of absorption distribution, the necessary conditions for convergence of the numerical solution to the analytical solution, form-factor prediction methods, and the computational requirements. The predictions are also used to investigate the effect of absorption distribution on sound fields in cubic enclosures with diffusely reflecting boundaries. Acoustical radiosity is then compared to predictions made in the four enclosures by a ray-tracing model that can account for diffuse reflection. Comparisons are made of echograms, room-acoustical parameters, and discretized echograms.

Effect of noise and occupancy on optimal reverberation times for speech intelligibility in classrooms.

The question of what is the optimal reverberation time for speech intelligibility in an occupied classroom has been studied recently in two different ways, with contradictory results. Experiments have been performed under various conditions of speech-signal to background-noise level difference and reverberation time, finding an optimal reverberation time of zero. Theoretical predictions of appropriate speech-intelligibility metrics, based on diffuse-field theory, found nonzero optimal reverberation times. These two contradictory results are explained by the different ways in which the two methods account for background noise, both of which are unrealistic. To obtain more realistic and accurate predictions, noise sources inside the classroom are considered. A more realistic treatment of noise is incorporated into diffuse-field theory by considering both speech and noise sources and the effects of reverberation on their steady-state levels. The model shows that the optimal reverberation time is zero when the speech source is closer to the listener than the noise source, and nonzero when the noise source is closer than the speech source. Diffuse-field theory is used to determine optimal reverberation times in unoccupied classrooms given optimal values for the occupied classroom. Resulting times can be as high as several seconds in large classrooms; in some cases, optimal values are unachievable, because the occupants contribute too much absorption.