Authors
Birgit Planitz, Paul Roe, Jiro Sumitomo, Michael Towsey, Ian Williamson, and Jason Wimmer (Queensland University of Technology)
Abstract
The need for large scale monitoring to manage environmental change is well established. Ecologists have long used acoustics to monitor faunal species but the scaling and automation of acoustic data gathering presents many problems. In this paper, we describe our steps to realising a complete acoustic environmental observatory – that is, an integrated set of hardware sensors, management utilities and analytical tools for large scale monitoring of the acoustic environment. Ecological scientists are collaborating with our system development.
We have developed several sensor types, including data loggers and autonomous wireless recording devices. We have successfully constructed and deployed sensors which utilise Smartphones to capture and transmit sound over a 3G network. When combined with a solar panel and batteries, we have maintained such sensors in remote locations for 6 months without human intervention.
We have developed a three step analysis sequence; energy analysis, event analysis and template analysis. Energy analysis locates acoustic energy in frequency and time and can be used to derive useful indicators of environmental health. Event analysis, which identifies energy events in a spectrogram that can be attributed to a single source, is a helpful complement to energy analysis.
Finally, template analysis adapts speech recognition and machine learning techniques to the recognition of specific faunal vocalizations. We have developed a web interface to support ecologists with analysis and annotation. Annotation is achieved using a folksonomy approach.
Recordings are simultaneously heard and viewed using a Microsoft Silverlight control. The web interface allows users to select a template to scan multiple recordings for presence of a pre-defined vocalisation. We present results which compare energy analysis with event analysis and results for template identification of bird calls. We conclude that automated acoustics has much to contribute to the investigation of species distributions and ecosystem health in general.
About the speaker
Professor Paul Roe received his MEng from the University of York in 1987 and his PhD from the University of Glasgow in 1991. He is currently a full professor in the Faculty of Science and Technology at QUT, in Brisbane Australia. At QUT he leads the Microsoft-QUT eResearch Centre, a collaboration between the Queensland State Government, Microsoft Research and QUT which is investigating smart tools for eResearch. Paul also chairs the QUT eResearch working party. He has published over 70 papers and received over $4M in competitive research funding. His research is focussed on smart tools which enable new forms and scales of research, particularly in the areas of distributed computing e.g. wireless sensors networks and end user programming languages.