callcut.pipeline.predict_recordings🔗

callcut.pipeline.predict_recordings(model, extractor, audio_paths, decoder, *, hop_frames=None)[source]🔗

Run inference on recordings and decode to call intervals.

For each recording: loads audio, extracts features, predicts frame-level probabilities, and decodes to call intervals. No ground truth annotations are needed.

Parameters:

modelBaseDetector: Trained model for call detection. Should already be on the desired device.
extractorBaseExtractor: Feature extractor matching the model’s expected input.
audio_pathslist of Path | str: Paths to audio files to process.
decoderBaseDecoder: Decoder for converting probabilities to call intervals.
hop_framesint | None: Hop between inference windows in frames. If None, uses the model’s default (75%% overlap).

Returns:

predictionslist of RecordingPrediction: Predicted call intervals for each recording.

Examples

>>> from callcut.pipeline import load_pipeline, predict_recordings
>>>
>>> model, extractor, decoder = load_pipeline("pipeline.pt")
>>> predictions = predict_recordings(model, extractor, audio_files, decoder)
>>> for pred in predictions:
...     print(f"{pred.audio_path.name}: {len(pred.intervals)} calls")