You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Recall: Proportion of actual voiced frames detected
F1: Harmonic mean of precision and recall
Pitch Accuracy Metrics
RMSE: Root Mean Square Error in Hz (lower is better)
Cents Error (Δ¢): Pitch deviation in musical cents (lower is better)
RPA: Raw Pitch Accuracy (higher is better)
RCA: Raw Chroma Accuracy (higher is better)
Combined
The overall performance of each algorithm is evaluated using a combined metric based on the harmonic mean of multiple key measurements. This approach was chosen specifically because:
Balanced Performance Requirement: The harmonic mean requires good performance across ALL metrics - an algorithm must perform well in both voicing detection and pitch accuracy to achieve a high score.
Penalty for Poor Performance: Poor performance in any single metric significantly impacts the final score, preventing algorithms from achieving high overall scores through excellence in just one area.
The combined metric incorporates four key measurements, all normalized to the range [0,1] where 1 is best: