CallTrackers uses three different automatic sound classifiers, managed by the British Trust for Ornithology (BTO), which can be accessed via the BTO Acoustic Pipeline desktop app. The classifiers search for audible calls from the Australasian bittern and Far eastern curlew and ultrasonic calls from Tasmanian bat species. Details on how to install and use the app to process CallTrackers recordings can be found in the User's Guides on the Documentation page.
All three classifiers produce spreadsheets that summarise the identifications, and the two bird classifiers also produce audio snippets of the identified calls for manual verification.
The classifiers are under development and can misidentify some sounds, so it's important to treat the results with caution. Classifiers are continually improving as more reference recordings are collected in our observations.
This page describes the current level of reliability of the three classifiers, which should be taken into account when examining the results.
The bittern and curlew classifiers are currently in development and register a high rate of false positives. Detections should be verified by listening to the original recordings.
The spreadsheet has one row per identification and shows the sound file name and a confidence "score". The confidence score is a number between 0 and 1 and is not a probability but a measure of how similar the recorded sound is to the training library of known calls: the higher the number the better. The training libraries for bitterns and curlews are still being developed, hence the need for manual verification.
The spreadsheet of identifications from the bittern and curlew classifiers should be considered as a short-list for follow-up listening rather than a list of definite identifications.
The bat classifier is also in development and performs better for some Tasmanian bat species than others. The table below summarises the reliability of the classifier for each species (with thanks to Stuart Newson at the BTO).
The spreadsheet produced by the BTO pipeline has one row per recorded file and a probability between 0 and 1. The probabilities relate to the false positive rate. For example, a probability of 0.7 (or 70%) means that on average we would expect 3 in 10 identifications to be assigned to the wrong species.
We can view and compare recorded bat calls to classified bat calls using freely available on-line resources and software. Wildlife Acoustics' Kaleidoscope software provides a means of viewing and listening to bat calls:
There are two on-line resources available to compare recordings to identified Tasmanian bat calls:
Tasmanian Bats and their Habitat - a Guide by Lisa Cawthen
The BTO's Ultrasonic Call Explorer, an on-line app which displays known calls for Tasmanian species and allows comparisons between species to be made.