Comparing manual and automated species recognition in the detection of four common south-east Australian forest birds from digital field recordings

作者：Joshi Kara A; Mulder Raoul A; Rowe Karen M C^*

来源：EMU, 2017, 117(3): 233-246.

DOI：10.1080/01584197.2017.1298970

• Automated recording unit
• automated recording device
• bioacoustics
• template-matching
• inventorying
• monitoring
• wildlife management
• conservation
• sound library
• Rhipidura fuliginosa
• Grey Fantail
• Cacomantis flabelliformis
• Fan-tailed Cuckoo
• Cormobates leucophaea
• White-throated Treecreeper
• Pardalotus punctatus
• Spotted Pardalote

Understanding and monitoring biological communities requires accurate species inventories using reliable and efficient methods of detection. Automated recorders deployed in the field have the potential to substantially reduce observer bias and the cost of human-based inventories, but their reliability and efficiency, particularly for common species of birds, are largely untested. Using digital audio data collected in 2013 by automated recorders within the Grampians National Park, Australia, we compared the performance of two species detection methods for four species of forest birds; manually scanning spectrograms and automating the identification of vocalisations by creating species-specific song templates, or 'recognisers', using the software program Song Scope (Wildlife Acoustics). Both detection methods generally identified the target species at the same sites, but species were detected at additional sites exclusively by one or the other method. Surprisingly, across species, manual scanning was more time-efficient overall than automated recognition, but results varied due to high rates of false positives and variable performance from the automated recognition recognisers we created. While automated recognition has the potential to improve species detection, the decision of whether to manually scan or automate recognition will depend on the species' vocalisation characteristics, calling frequency, and template-matching algorithms used.

• 出版日期2017