Computational Bioacoustics

Todor Ganchev

• Publisher: De Gruyter
• Year: 2017
• ISBN:
  9781614519669
• Edition: 1
• Topic: Tecnologia e ingegneria
• Subtopic: Ingegneria acustica

This book offers an overview of some recent advances in the Computational Bioacoustics methods and technology. In the focus of discussion is the pursuit of scalability, which would facilitate real-world applications of different scope and purpose, such as wildlife monitoring, biodiversity assessment, pest population control, and monitoring the spread of disease transmitting mosquitoes. The various tasks of Computational Bioacoustics are described and a wide range of audio parameterization and recognition tasks related to the automated recognition of species and sound events is discussed. Many of the Computational Bioacoustics methods were originally developed for the needs of speech, audio, or image processing, and afterwards were adapted to the requirements of automated acoustic recognition of species, or were elaborated further to address the challenges of real-world operation in 24/7 mode. The interested reader is encouraged to follow the numerous references and links to web resources for further information and insights. This book is addressed to Software Engineers, IT experts, Computer Science researchers, Bioacousticians, and other practitioners concerned with the creation of new tools and services, aimed at enhancing the technological support to Computational Bioacoustics applications.

1 Introduction

1.1 Why should we care about biodiversity?

Biodiversity is essential to the health of Earth’s ecosystems. The consequences of biodiversity loss translate to impoverishment of environment, lower productivity, and lessening its capacity to support complex life, including the human species. The improved public awareness about these consequences made possible the prioritization of efforts aimed at the development of conservation planning policies. The last one brought forward the need of ecological impact assessment of all human activities, which indirectly contributes to the biodiversity preservation efforts (Rands et al. 2010; Collen et al. 2013).

Essentially, in the past decade, biodiversity preservation efforts gained some wider public and governmental support.3, 4 This is a good indicator of the awakened public consciousness about the seriousness of these problems, and the comprehension of associated risks and potentially devastating consequences on a global scale. If these risks are not managed appropriately and in a timely manner, further escalation of threat might pass the point of no return, which will trigger a chain of global catastrophic events that will probably affect the entire human civilization.

To this end, it has been undoubtedly proven that the efforts invested in biodiversity preservation, conservation planning, and environmental impact assessment bring positive effects to all human economic activities and contribute to improving the quality of life. In the same time, however, at present, the level of public support is limited to specific local problems and therefore is not always aimed at solving the global problems of biodiversity preservation.

This is because global biodiversity monitoring initiatives, such as the Global Biodiversity Information Facility (GBIF),5 typically require monitoring of various families of animal species, inhabiting large territories, in order to assess the practical effects of certain measures and mitigation actions implemented in support of biodiversity conservation.

Therefore, the monitoring of certain key biodiversity indicators is a mandatory task of every effort in support of biodiversity preservation. Habitat protection and biodiversity preservation activities are profoundly dependent on the availability of rapid and precise biodiversity assessment survey methods and on the feasibility of continuous long-term monitoring of certain biodiversity indicators (Hill et al. 2005). These are mandatory components of every conservation action; however, neither of these are trivial tasks! Fulfilling these and other related tasks is challenging mostly because nowadays biodiversity assessment studies depend on the involvement of human experts on all stages of work. By that reason, biodiversity assessment and biodiversity monitoring actions are time demanding, expensive, and therefore not scalable.

The high cost of expert-based studies, and the enormous time effort required for the implementation of biodiversity monitoring and biodiversity assessment surveys, imposes limitations on their scope. By that reason, such studies are implemented only periodically, predominantly during daytime and favourable weather conditions, in locations that are less challenging logistically, and so on. These and other practical limitations currently narrow the spatial and temporal coverage of biodiversity-related studies and impede the efforts to understand the overall biodiversity dynamics on a global scale. All this significantly limits our chances to comprehend and model the underlying large-scale processes and to figure out and implement efficient conservation strategies.

Hence, it is highly desirable to provide strong technological support to biodiversity monitoring and assessment actions. Automation of certain tedious biodiversity monitoring procedures and ecological impact assessment efforts is seen as the only reasonable approach for addressing successfully the above-mentioned challenges imposed by long-term large-scale efforts aiming at biodiversity preservation.

Fortunately, technological tools and automation of data collection procedures started to gain trust in biodiversity assessment studies and were reported to enhance significantly the scope and temporal coverage of these surveys when compared to traditional human experts-based surveys. In that sense, it is already acknowledged that a well-designed technological support could complement the work of human experts, and thus, facilitate the global efforts to preserve biodiversity and the vitality of life-supporting environment on the Earth.

1.2 Role of computational bioacoustics6

Sounds play an important role in nature. For instance, animals use sounds for communication in order to recognize and localize potential prey, to repel contenders, or to evade predators. Besides, sounds are emitted as a by-product of typical activities, such as locomotion, feeding, and so on. Therefore, animal sounds are considered an important information source in biodiversity assessment studies – sound propagates well in darkness, and when compared to eyesight is less affected by vegetation and obstructions. By that reason, most often animals are easier to hear than to see (e. g. due to their small size, camouflage, nocturnal activity patterns, etc.). These and other reasons established bioacoustic methods as major survey means and essential data contributors in biodiversity assessment studies (Pijanowski et al. 2011a, 2011b).

Terrestrial bioacoustic studies typically rely on passive monitoring methods, which remotely register the acoustic emissions of animals in their natural habitats.7 The greatest advantage of passive bioacoustic methods is that they are less obtrusive when compared to traditional study methods, such as mark recapture, GPS tracking, mist netting, DNA analysis, and so on (Hill et al. 2005). Audio-visual survey methods (Jahn 2011) obviate the negative consequences of animal capture and any physical contact with animals, which helps to evade risks related to accidental death of captured animals. However, the audio-visual survey methods are human expert based, and therefore, their successful application depends on the availability of highly skilled biologists and their extended presence on the field. The last does not facilitate long-term studies and scalability. Important advantages of passive bioacoustic methods are that

1. they are remote, that is, animals can be observed from a distance;

2. they do not cause significant disturbance to animal behaviours;

3. they do not depend on physical contact with the animals;

4. certain steps of data collection, management, and processing can be automated with the use of contemporary technology; and

5. observational data are recorded and can be stored for prolonged periods of time, analysed at a later stage, repeatedly used in long-term comparative studies.

To this end, bioacoustics already makes extensive use of technological tools. For instance, nowadays, autonomous recording devices are capable of collecting data over prolonged periods (weeks and months) without maintenance for battery or storage replacement. Furthermore, certain brands of equipment also possess the functionality to transmit wirelessly all, or some portion of, the recorded data in near real time or on demand. However, the data processing, analysis, and interpretation tasks largely depend on the availability (and the qualification) of experienced bioacousticians who need to inspect and/or listen to a certain amount of audio recordings.

The main point here is that although nowadays technology provides relatively cheap ways of collecting continuous recordings in various habitats, the workflow of data analysis and interpretation is high priced. This is because bioacousticians need to manually inspect and process all recordings (or in the best case in a computer-assisted manner) in order to carry out most of the tasks related to data analysis and interpretation. At the same time, a large-scale biodiversity monitoring project would require continuous recording over multiple locations, and therefore, the scale of effort and cost required becomes prohibitive – primarily due to the big data problem and the limited scalability of human expert-based study methods. The limited scalability is seen as the major obstacle that restricts the scope and time scale of biodiversity monitoring actions. In a typical present-day study, these are restricted to few particular species, monitored periodically in a relatively small area, over a number of short fieldwork missions.

In order to explain better the role of computational bioacoustics, in Chapter 2, we provide details on the historical roots of bioacoustics, briefly account to the transformation of needs, tasks, scope, and methods during its exciting evolution over the past 100 years, and then discuss its ongoing transformation to computational bioacoustics. In brief, the remarkable advance of information processing methods and communication technology in the past decade created the prerequisites for the emergence of a wide range of automated tools for data acquisition, transmission, storage, processing, and visualization. The conjunction of traditional bioacoustics with advanced information extraction and knowledge retrieval methods, communications technology, computer science resulted in the emergence of the new scientific discipline –computational bioacoustics.

Computational bioacoustics aims at scalable biodiversity monitoring. Therefore, it aims to develop automated methods and tools that provide the badly needed technological support so that problems related to big data acquisition, organization, processing, analysis, and interpretation are addressed in a scalable manner.

In the present book, we focus on topics of computational bioacoustics that are closely related to contemporary methods for automated acoustic recognition of species and sound events. These methods comprise the fundamentals of technological support to real-world applications, aiming at automated biodiversity monitoring and assessment, pest control, monitoring the spread of disease-transmitting insects, and other related applications where automated acoustic recognition of species is of primary concern. In Chapter 2, we briefly outline some applications where automation of sound recognition conveys the biggest impact to the current expert-based methods, and in Chapter 3, we define the main technological tasks that are in the focus of discussion throughout subsequent chapters.

1.3 The benefits of computational bioacoustics

Computational bioacoustics aims at scalable solutions, which facilitate the implementation of large-scale biodiversity monitoring studies. In the long term, these solutions contribute to the creation of advanced audio processing technology in support of successful coping with the global challenges of biodiversity preservation.

Computational bioacoustics aims to develop robust audio processing methods. Taking advantage of the fast progress of information technology, computational bioacoustics, among other tasks, aims to develop robust audio processing methods that successfully cope with the challenges of real-world operation and provide scalable tools for the benefit of biodiversity assessment and conservation actions. These methods and tools constitute the foundations of automated technological support that would permit scalability and would make feasible the implementation of unattended long-term biodiversity monitoring, acoustic pest control, audio information retrieval, and other relevant services for data analysis and interpretation. For instance, automated acoustic monitoring tools provide the means for capturing acoustic activity of species, which are rarely seen due to nocturnal activity or which occupy hardly accessible habitats, dangerous places,8 and so on. Among the most important advantages of passive automated acoustic monitoring is that it is unobtrusive to animals and can be implemented continuously in 24/7 mode over extended periods of time. In this regard, computational bioacoustics well supports the concept of integrated archival of raw recordings, audio processing tools, the corresponding statistical models, and data analysis results. A great advantage of the computational bioacoustics methods comes from the fact that digital audio recordings collected by automated recording stations can be duly archived over extended periods of time, and when needed rechecked and reprocessed multiple times with different tools. In such a way, together with the original and processed data, one has the opportunity to store also the software tools with the specific settings used to obtain these results, together with the models created, and together with the outcome of the data analysis. Such a functionality facilitates reproducible research, offers opportunities for a signif...