After viewing our Kaleidoscope tutorials, it can be tempting to build advanced classifiers for every species in your area using small batches of training data and using those to search through a much larger dataset. However, for applications that require a general survey of all recorded species, the most efficient option is more straightforward.
We recommend simply performing simple clustering on your full dataset, then going through each cluster and using File > Bulk ID to label each cluster by species. In ambiguous clusters or ones that contain multiple species, you can instead manually identify individual vocalizations.
It is worth noting that whether you perform simple clustering or build classifiers from individually labeled vocalizations, any form of algorithmic pattern recognition will produce false positives and false negatives and will require verification by a human expert for applications that call for full accuracy.