VoxShield returns a confidence-related value alongside the predicted label.
For tree-based classifiers that expose predict_proba(), this value represents the model's estimated class probability under its learned model.
predict_proba()
It should not automatically be interpreted as a calibrated probability of authenticity.
Example:
{ "result": "FAKE", "confidence": 0.87 }
Interpretation
The example indicates a model output associated with the FAKE class.
It does not prove that the audio is cloned with 87% real-world certainty.
Recommended Documentation
The confidence output should be described as a model confidence estimate, unless calibration and reliability have been explicitly evaluated.
There was an error while loading. Please reload this page.