It's still a bit early days. For nocturnal flight calls (which are the majority of the mysteries) the recordings are generally of poor quality and the model doesn't find anything (according to the tresholds we have set). We are running the model in the background on focal recordings from Europe as well, and there performance is pretty good. As we are not runnign BirdNET alongside AvesEcho, I cannot say at this stage which one is better, but my expectation is that AvesEcho should be an improvement, as it is based on BirdNET and is then further trained with more focal recordings. But time will tell if that is indeed the case!