In a nutshell, in the portal you are asked to identify and label the species that vocalize in a ten-second audio recording. Labelled boxes (Fig. 1) will be used for training the models and identifications without boxes will be used for evaluating the performance of these models. Already some dozens to some hundreds of labelled vocalizations per species can significantly improve the model performance in a given location.
The end results will be used to identify bird sounds collected as part of the LIFEPLAN research programme, which aims to explain global biodiversity and its driving factors. Produced identification models will also be made openly available for everyone.
Do you have bioacoustics data of birds, and are you interested in automated sound identification? BSG hosts audio data from 80 sites collected as part of the LIFEPLAN research programme, but more audio data can be imported to the web portal. Groups who have provided said bioacoustics data can then annotate it. In exchange for the data and annotations, BSG will provide automated bird sound identification models made from said data.
Please contact Sebastian Andrejeff ([email protected]) if you have any questions or are interested in annotating your bird bioacoustics data and getting automated identification models based on it.