Evaluate

bt4vt.evaluate.compute_fpfnth(scores, labels)

Calculation of False Positive Rates and False Negative Rates and corresponding thresholds

Parameters:

• scores (pandas.Series) – Series of scores

• labels (pandas.Series) – Series of labels; labels have to be either {-1,1} or {0,1}

Returns:

fprs, fnrs, thresholds

Return type:

ndarray, ndarray, ndarray

bt4vt.evaluate.evaluate_scores(scores, labels, dcf_costs, threshold_values=None)

Evaluation of scores for the overall dataset and for specified speaker groups. In the average case no threshold_values are provided. Threshold values are used to compute the detection cost function for specified speaker groups. The function returns False Positive Rates, False Negative Rates and corresponding thresholds as well as the corresponding metric scores. In the average case, metric thresholds are returned in addition.

Parameters:

• scores (pandas.Series) – Series of scores

• labels (pandas.Series) – Series of labels

• dcf_costs (list) – list of tuples specifying the weights for the detection cost function (dcf_p_target, dcf_c_fp, dcf_c_fn)

• threshold_values (pandas.Series) – Series of threshold values computed for the overall dataset and used to determine the metric scores for the specified speaker groups

Returns:

fprs, fnrs, thresholds, metric_scores, (metric_thresholds)

Return type:

ndarray, ndarray, ndarray, list, (list)