Birdnet Modular cache

[nishantnnb]

hi,
Created this repository for modular cache. It can create species-wise cache, combine and split the cache. I believe, it can help reducing the time required during the end-to-end training / re-training pipeline. Probable use cases:

1. Fragment the load of data preparation: Different team members can prepare NPZ files for different species and combine them later into one training set. This load fragmentation can save time during the training process.

2. Model retraining: If only a few species need more samples or updated negatives, only those species caches need to be replaced, then all caches can be combined again. This is substantial time saving while retraining a model.

3. Collaboration: Species-wise cache files are easier to exchange, and collaborators can share embeddings instead of raw recordings.

4. Model upgrade: If a model already covers 100 species and 20 more must be added, only the new 20 species need fresh cache preparation before combining with the old set.

https://github.com/nishantnnb/Birdnet_modular_cache/blob/main/README.md

Regards,
Nishant B.