Thanks to audio recorders with microphones capable of recording bat calls, we can have a lot of recordings. It is a positive situation, but the problem is how to process this data. Software such as Kaleidoscope costs hundreds of Euros, but visual control of spectrograms is almost impossible for thousands of recordings. Today, only a few open-source alternatives exist. And one is Batdetect2.
Description
I have used Batdetect2 for analysing recordings from automatic monitoring recorders. Especially the detection capability is great; almost every call is detected. This model is trained on data for only 17 European bat species. However, if it does not recognize the call, it marks it as the species that is most similar to the trained sample. My experience is that problematic species are Eptesicus nilssonii, Eptesicus isabellinus, Hypsugo savii, Miniopterus schreibersii, Myotis dasycneme, Nyctalus lasiopterus, Pipistrellus kuhlii in Slovakia. Yes, many species from the Myotis genus are also problematic, but it is almost impossible to distinguish these species based on spectrogram analysis. I would appreciate training the model for Hypsugo savii, Nyctalus lasiopterus and Myotis dasycneme (long quail-like calls produced over water surfaces).
Notes to identification issues
The model has some issues with certain specific calls, which I noticed when social calls of Pipistrellus pygmaeus were present. Many of them were identified as Plecotus auritus, Nyctalus noctula, or Eptesicus serotinus calls with a low level of classification probability (typically around 0.1). Sometimes "weird" sounds, probably from insects, are interpreted as bat calls — Nyctalus/Eptesicus or Pipistrellus. Almost in all cases when the model identified only one bat call in a 1-minute recording, it was a false identification. Thus, Batdetect2 is a useful tool for identifying a few species and for detecting bat calls. However, the main part of the signals has to be verified by an expert.