Custom Confidence Labeling

[pastorep]

Currently, our Era 1 data generation puts k-ranking and confidence score in the output .wav filenames (i.e. k=1 is clip with highest confidence, k=2 is next most confident, ...). However, these confidence scores come from the existing SRKW detection model at the time of annotation. It would be nice to allow plugin-style logic to label the files with whatever confidence and k-value one would like. Doing so allows us to decouple saved clips from the confidence scores assigned to them at time of annotation.

Use Case: I run my own custom model on every detection clip (1sec) in the detection segment (60sec) and label the generated .wav files with my custom model's k-rank/confidence in true label.