Metrics

bt4vt.metrics.compute_cdet_at_threshold(fprs, fnrs, thresholds, threshold_value, dcf_p_target, dcf_c_fp, dcf_c_fn)[source]

Computation of detection cost function at a given threshold. Computation is performed as defined in the NIST Speaker Recognition Evaluation Plan 2019

\[C_{Det}(\theta) = C_{FN} \times P_{Target} \times P_{FN}(\theta) + C_{FP} \times (1 - P_{Target}) \times P_{FP}(\theta)\]
Parameters:

  • fprs (ndarray) – Array of False Positive Rates

  • fnrs (ndarray) – Array of False Negative Rates

  • thresholds (ndarray) – Array of Threshold values corresponding to fprs and fnrs

  • threshold_value (float) – Threshold value to compute detection cost function for

  • dcf_p_target (float) – [description]

  • dcf_c_fp (float) – [description]

  • dcf_c_fn (float) – [description]

Returns:

cdet_at_threshold

Return type:

float

bt4vt.metrics.compute_eer(fprs, fnrs, thresholds)[source]

Computation of the Equal Error Rate and its threshold

Parameters:

  • fprs (ndarray) – Array of False Positive Rates

  • fnrs (ndarray) – Array of False Negative Rates

  • thresholds (ndarray) – Array of Threshold values corresponding to fprs and fnrs

Returns:

eer, eer_threshold

Return type:

float, float

bt4vt.metrics.compute_metrics_ratios(metrics)[source]

Computation of metric ratios defined as the subgroup metric scores divided by the average metric score.

Parameters:

metrics (DataFrame) – DataFrame that contains metric scores for the average evaluation and subgroup evaluations. The first row corresponds to the eer, all other rows correspond to the min_cdet scores with weights specified in the config file

Returns:

metric_ratios

Return type:

DataFrame

bt4vt.metrics.compute_min_cdet(fprs, fnrs, thresholds, dcf_p_target, dcf_c_fp, dcf_c_fn)[source]

Computation of the minimum of the detection cost function and its threshold. Computation is performed as defined in the NIST Speaker Recognition Evaluation Plan 2019

\[C_{Det}(\theta) = C_{FN} \times P_{Target} \times P_{FN}(\theta) + C_{FP} \times (1 - P_{Target}) \times P_{FP}(\theta)\]
Parameters:

  • fprs (ndarray) – Array of False Positive Rates

  • fnrs (ndarray) – Array of False Negative Rates

  • thresholds (ndarray) – Array of Threshold values corresponding to fprs and fnrs

  • dcf_p_target (float) – [description]

  • dcf_c_fp (float) – [description]

  • dcf_c_fn (float) – [description]

Returns:

min_cdet, min_cdet_threshold

Return type:

float, float

bt4vt.metrics.get_fpfn_at_threshold(fprs, fnrs, thresholds, threshold_value, ppf_norm=False)[source]

Get the False Positive Rate and False Negative Rate at a given threshold value.

Parameters:

  • fprs (ndarray) – Array of False Positive Rates

  • fnrs (ndarray) – Array of False Negative Rates

  • thresholds (ndarray) – Array of Threshold values corresponding to fprs and fnrs

  • threshold_value (float) – Threshold value to get fpr and fnr for i.e. min_cdet_threshold

  • ppf_norm (bool) – normalise the fpr and fnr values to the percent point function. Default is set to False.

Returns:

fpr_at_threshold, fnr_at_threshold

Return type:

float, float