An evaluation of manual and automated methods for detecting sounds of maned wolves (Chrysocyon brachyurusIlliger 1815)

Abstract

Although bioacoustics is increasingly used to study species and environments for their monitoring and conservation, detecting calls produced by species of interest is prohibitively time consuming when done manually. Here we compared four methods for detecting and identifying roar-barks of maned wolves (Chrysocyon brachyurus) within long sound recordings: (1) a manual method, (2) an automated detector method using Raven Pro 1.4, (3) an automated detector method using XBAT and (4) a mixed method using XBAT's detector followed by manual verification. Recordings were done using a song meter installed at the Serra da Canastra National Park (Minas Gerais, Brazil). For each method we evaluated the following variables in a 24-h recording: (1) total time required analysing files, (2) number of false positives identified and (3) number of true positives identified compared to total number of target sounds. Automated methods required less time to analyse the recordings (77 -93 min) when compared to manual method (189 min), but consistently presented more false positives and were less efficient in identifying true positives (manual ¼ 91.89%, Raven ¼ 32.43% and XBAT ¼ 84.86%). Adding a manual verification after XBAT detection dramatically increased efficiency in identifying target sounds (XBAT þ manual ¼ 100% true positives). Manual verification of XBAT detections seems to be the best way out of the proposed methods to collect target sound data for studies where large amounts of audio data need to be analysed in a reasonable time (111 min, 58.73% of the time required to find calls manually).

Automated birdsong recognition in complex acoustic environments: a review

“…Therefore, the majority of the published work has largely avoided automatic segmentation and instead used manually segmented data to test their recognition methods (Anderson et al 1996, Franzen and Gu 2003, Chen and Maher 2006, Somervuo et al 2006, Fox et al 2008. Even when automatic segmentation is used, it is still followed by manual elimination of noisy segments (Selin et al 2007, Rocha et al 2015.…”

“…While there has been considerable research on automated birdsong recognition to date, the need for further development is evident given that ecologists and conservation managers still spend a great deal of time manually scanning field recordings because none of the available automated birdsong recognition software fulfil their requirements reliably (Swiston and Mennill 2009, Goyette et al 2011, Potamitis 2014, Ulloa et al 2016). However, the potential of automated detection combined with a degree of human calibration to cope with large datasets is encouraging (Urazghildiiev and Clark 2007, Digby et al 2013, Brighten 2015, Rocha et al 2015.…”

Bioacoustic monitoring of animal vocal behavior for conservation

“…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). For conservation programs aiming to detect rare or cryptic species or behaviors, specificity should be low.…”

Upcall production by southern right whale (Eubalaena australis) mother‐calf pairs may be independent of diel period in a nursery area

“…); and (3) a combined technique using the NARW automated tool associated with manual browsing of 1 min segments before and after each detection event using XBAT (Figueroa ) as described in Rocha et al . .…”

“…(2) an automated detection tool developed for NARW upcalls Clark 2006, Urazghildiiev et al 2009); and (3) a combined technique using the NARW automated tool associated with manual browsing of 1 min segments before and after each detection event using XBAT (Figueroa 2007) as described in Rocha et al 2015.…”