Pipeline

callcut.pipeline provides end-to-end evaluation, prediction, and serialization for trained call detection pipelines.

Evaluation and prediction

evaluate_recordings() runs a full evaluation pipeline on annotated recordings: inference, decoding, interval matching, and metric computation. Use it with test_recordings to evaluate on held-out data after training.

predict_recordings() runs inference on new audio files without ground truth.

evaluate_recordings(model, extractor, ...[, ...])	Evaluate a trained model on annotated recordings.
predict_recordings(model, extractor, ...[, ...])	Run inference on recordings and decode to call intervals.

Serialization

save_pipeline() saves a complete pipeline (model, extractor, decoder) to a single file. load_pipeline() reconstructs all components from the saved file.

save_pipeline(model, extractor, decoder, ...)	Save a complete pipeline to a file.
load_pipeline(fname, *[, device])	Load a complete pipeline from a file.

Result types

EvaluationReport(recordings, event_metrics, ...)	Aggregate evaluation results across multiple recordings.
RecordingEvaluation(recording, ground_truth, ...)	Evaluation results for a single recording.
RecordingPrediction(audio_path, intervals, ...)	Predicted call intervals for a single recording.