Automated birdsong recognition in complex acoustic environments: a review

Abstract

Conservationists are increasingly using autonomous acoustic recorders to determine the presence/absence and the abundance of bird species. Unlike humans, these recorders can be left in the field for extensive periods of time in any habitat. Although data acquisition is automated, manual processing of recordings is labour intensive, tedious, and prone to bias due to observer variations. Hence automated birdsong recognition is an efficient alternative. However, only few ecologists and conservationists utilise the existing birdsong recognisers to process unattended field recordings because the software calibration time is exceptionally high and requires considerable knowledge in signal processing and underlying systems, making the tools less user‐friendly. Even allowing for these difficulties, getting accurate results is exceedingly hard. In this review we examine the state‐of‐the‐art, summarising and discussing the methods currently available for each of the essential parts of a birdsong recogniser, and also available software. The key reasons behind poor automated recognition are that field recordings are very noisy, calls from birds that are a long way from the recorder can be faint or corrupted, and there are overlapping calls from many different birds. In addition, there can be large numbers of different species calling in one recording, and therefore the method has to scale to large numbers of species, or at least avoid misclassifying another species as one of particular interest. We found that these areas of importance, particularly the question of noise reduction, are amongst the least researched. In cases where accurate recognition of individual species is essential, such as in conservation work, we suggest that specialised (species‐specific) methods of passive acoustic monitoring are required. We also believe that it is important that comparable measures, and datasets, are used to enable methods to be compared.

Bioacoustic monitoring of animal vocal behavior for conservation

“…Performance depends in part on extraneous sources of sound (e.g., other species' calls) and the overall noisiness of the environment (e.g., anthropogenic noise, wind, rain), as well as the acoustic structure of the vocalizations, which is important in the choice of algorithm (Brandes, 2008;Cragg, Burger, & Piatt, 2015;Priyadarshani, Marsland, & Castro, 2018;Salamon et al, 2016;Towsey, Planitz, Nantes, Wimmer, & Roe, 2012). Since bioacoustic monitoring typically generates very large volumes of sound data, algorithms to detect vocalizations from sound files, termed call recognizers, are critical to the success of bioacoustics as a wildlife monitoring tool.…”

“…This includes the often-variable calls of juveniles (Priyadarshani et al, 2018). If the species being monitored has a large repertoire, researchers or practitioners using bioacoustic methods should determine which vocalizations are of most use considering the program's aims, and tailor the recognizer towards these.…”

“…In such cases, manual verification will most likely be necessary, the time commitments for which must be considered (Cragg et al, 2015;Rocha, Ferreira, Paula, Rodrigues, & Sousa-Lima, 2015). Moreover, packages like monitoR, as well as other commercial software, assist enormously in making recognizer development accessible to the non-expert (i.e., people without expert programming skills), which is crucial if bioacoustics is to become widely applied in threatened species monitoring (Priyadarshani et al, 2018;Sebastián-González et al, 2015). For this reason, recognizers should ideally be built to align with the project's aims (Priyadarshani et al, 2018), or should be easily altered as needs be.…”

Koe: Web‐based software to classify acoustic units and analyse sequence structure in animal vocalizations

“…producing a ‘catalogue' of class types) and then assigning units to those classes; only once units have been classified can repertoires and sequence structures be analysed and compared. Classification is a challenge, however; thus far, attempts at automation have not proven generalizable (Priyadarshani, Marsland, & Castro, ) and the primary approach for most species remains manual classification based on human visual and auditory perception (Kershenbaum et al, ).…”

AviaNZ: A future‐proofed program for annotation and recognition of animal sounds in long‐time field recordings

“…Although mathematicians, computer scientists and engineers have started to work on methods for the analysis of these data, it has proven to be a challenging topic, particularly once the need for high levels of accuracy in identification is considered (Priyadarshani, Marsland, & Castro, ). In this paper, we describe our open‐source software that aims to facilitate this.…”